- Split bearch.h correctly into bearch.h and bearch_t.h
[libfirm] / ir / be / ia32 / ia32_emitter.c
index 0a549cb..68905ea 100644 (file)
@@ -1,9 +1,9 @@
 /**
- * This file implements the node emitter.
- * @author Christian Wuerdig
- * $Id$
+ * @file
+ * @brief   This file implements the node emitter.
+ * @author  Christian Wuerdig, Matthias Braun
+ * @version $Id$
  */
-
 #ifdef HAVE_CONFIG_H
 #include "config.h"
 #endif
 #include "irargs_t.h"
 #include "irprog_t.h"
 #include "iredges_t.h"
+#include "execfreq.h"
+#include "error.h"
+#include "raw_bitset.h"
 
 #include "../besched_t.h"
 #include "../benode_t.h"
+#include "../beabi.h"
+#include "../be_dbgout.h"
+#include "../beemitter.h"
+#include "../begnuas.h"
+#include "../beirg_t.h"
 
 #include "ia32_emitter.h"
 #include "gen_ia32_emitter.h"
 #include "ia32_map_regs.h"
 #include "bearch_ia32_t.h"
 
-#define BLOCK_PREFIX(x) ".L" x
-
-#define SNPRINTF_BUF_LEN 128
-
-/* global arch_env for lc_printf functions */
-static const arch_env_t *arch_env = NULL;
-
-/** by default, we generate assembler code for the Linux gas */
-asm_flavour_t asm_flavour = ASM_LINUX_GAS;
-
-/**
- * Switch to a new section
- */
-void ia32_switch_section(FILE *F, section_t sec) {
-       static section_t curr_sec = NO_SECTION;
-       static const char *text[ASM_MAX][SECTION_MAX] = {
-               {
-                       ".section\t.text", ".section\t.data", ".section\t.rodata", ".section\t.text"
-               },
-               {
-                       ".section\t.text", ".section\t.data", ".section .rdata,\"dr\"", ".section\t.text"
-               }
-       };
-
-       if (curr_sec == sec)
-               return;
-
-       curr_sec = sec;
-       switch (sec) {
-
-       case NO_SECTION:
-               break;
+DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
 
-       case SECTION_TEXT:
-       case SECTION_DATA:
-       case SECTION_RODATA:
-       case SECTION_COMMON:
-               fprintf(F, "\t%s\n", text[asm_flavour][sec]);
-       }
-}
-
-static void ia32_dump_function_object(FILE *F, const char *name)
-{
-       switch (asm_flavour) {
-       case ASM_LINUX_GAS:
-               fprintf(F, "\t.type\t%s, @function\n", name);
-               break;
-       case ASM_MINGW_GAS:
-               fprintf(F, "\t.def\t%s;\t.scl\t2;\t.type\t32;\t.endef\n", name);
-               break;
-       }
-}
-
-static void ia32_dump_function_size(FILE *F, const char *name)
-{
-       switch (asm_flavour) {
-       case ASM_LINUX_GAS:
-               fprintf(F, "\t.size\t%s, .-%s\n", name, name);
-               break;
-       }
-}
-
-/*************************************************************
- *             _       _    __   _          _
- *            (_)     | |  / _| | |        | |
- *  _ __  _ __ _ _ __ | |_| |_  | |__   ___| |_ __   ___ _ __
- * | '_ \| '__| | '_ \| __|  _| | '_ \ / _ \ | '_ \ / _ \ '__|
- * | |_) | |  | | | | | |_| |   | | | |  __/ | |_) |  __/ |
- * | .__/|_|  |_|_| |_|\__|_|   |_| |_|\___|_| .__/ \___|_|
- * | |                                       | |
- * |_|                                       |_|
- *************************************************************/
-
-static INLINE int be_is_unknown_reg(const arch_register_t *reg) {
-       return \
-               REGS_ARE_EQUAL(reg, &ia32_gp_regs[REG_GP_UKNWN])   || \
-               REGS_ARE_EQUAL(reg, &ia32_xmm_regs[REG_XMM_UKNWN]) || \
-               REGS_ARE_EQUAL(reg, &ia32_vfp_regs[REG_VFP_UKNWN]);
-}
-
-/**
- * returns true if a node has x87 registers
- */
-static INLINE int has_x87_register(const ir_node *n) {
-       return is_irn_machine_user(n, 0);
-}
-
-/* We always pass the ir_node which is a pointer. */
-static int ia32_get_arg_type(const lc_arg_occ_t *occ) {
-       return lc_arg_type_ptr;
-}
+#define BLOCK_PREFIX ".L"
 
+#define SNPRINTF_BUF_LEN 128
 
 /**
  * Returns the register at in position pos.
  */
-static const arch_register_t *get_in_reg(const ir_node *irn, int pos) {
+static
+const arch_register_t *get_in_reg(ia32_emit_env_t *env, const ir_node *irn,
+                                  int pos)
+{
+       const arch_env_t       *arch_env = env->arch_env;
        ir_node                *op;
        const arch_register_t  *reg = NULL;
 
@@ -142,13 +67,22 @@ static const arch_register_t *get_in_reg(const ir_node *irn, int pos) {
 
        assert(reg && "no in register found");
 
-       /* in case of unknown: just return a register */
-       if (REGS_ARE_EQUAL(reg, &ia32_gp_regs[REG_GP_UKNWN]))
-               reg = &ia32_gp_regs[REG_EAX];
-       else if (REGS_ARE_EQUAL(reg, &ia32_xmm_regs[REG_XMM_UKNWN]))
-               reg = &ia32_xmm_regs[REG_XMM0];
-       else if (REGS_ARE_EQUAL(reg, &ia32_vfp_regs[REG_VFP_UKNWN]))
-               reg = &ia32_vfp_regs[REG_VF0];
+       /* in case of a joker register: just return a valid register */
+       if (arch_register_type_is(reg, joker)) {
+               const arch_register_req_t *req;
+
+               /* ask for the requirements */
+               req = arch_get_register_req(arch_env, irn, pos);
+
+               if (arch_register_req_is(req, limited)) {
+                       /* in case of limited requirements: get the first allowed register */
+                       unsigned idx = rbitset_next(req->limited, 0, 1);
+                       reg = arch_register_for_index(req->cls, idx);
+               } else {
+                       /* otherwise get first register in class */
+                       reg = arch_register_for_index(req->cls, 0);
+               }
+       }
 
        return reg;
 }
@@ -156,9 +90,13 @@ static const arch_register_t *get_in_reg(const ir_node *irn, int pos) {
 /**
  * Returns the register at out position pos.
  */
-static const arch_register_t *get_out_reg(const ir_node *irn, int pos) {
-       ir_node                *proj;
-       const arch_register_t  *reg = NULL;
+static
+const arch_register_t *get_out_reg(ia32_emit_env_t *env, const ir_node *irn,
+                                   int pos)
+{
+       const arch_env_t      *arch_env = env->arch_env;
+       ir_node               *proj;
+       const arch_register_t *reg = NULL;
 
        /* 1st case: irn is not of mode_T, so it has only                 */
        /*           one OUT register -> good                             */
@@ -167,11 +105,9 @@ static const arch_register_t *get_out_reg(const ir_node *irn, int pos) {
 
        if (get_irn_mode(irn) != mode_T) {
                reg = arch_get_irn_register(arch_env, irn);
-       }
-       else if (is_ia32_irn(irn)) {
+       } else if (is_ia32_irn(irn)) {
                reg = get_ia32_out_reg(irn, pos);
-       }
-       else {
+       } else {
                const ir_edge_t *edge;
 
                foreach_out_edge(irn, edge) {
@@ -188,197 +124,264 @@ static const arch_register_t *get_out_reg(const ir_node *irn, int pos) {
        return reg;
 }
 
-enum io_direction {
-  IN_REG,
-  OUT_REG
-};
-
 /**
- * Returns the name of the in register at position pos.
+ * Returns an ident for the given tarval tv.
  */
-static const char *get_ia32_reg_name(ir_node *irn, int pos, enum io_direction in_out) {
-       const arch_register_t *reg;
-
-       if (in_out == IN_REG) {
-               reg = get_in_reg(irn, pos);
+static
+ident *get_ident_for_tv(tarval *tv) {
+       char buf[256];
+       tarval_snprintf(buf, sizeof(buf), tv);
+       return new_id_from_str(buf);
+}
 
-               if (reg->reg_class == &ia32_reg_classes[CLASS_ia32_vfp]) {
-                       /* FIXME: works for binop only */
-                       assert(2 <= pos && pos <= 3);
-                       reg = get_ia32_attr(irn)->x87[pos - 2];
+/**
+ * Determine the gnu assembler suffix that indicates a mode
+ */
+static
+char get_mode_suffix(const ir_mode *mode) {
+       if(mode_is_float(mode)) {
+               switch(get_mode_size_bits(mode)) {
+               case 32:
+                       return 's';
+               case 64:
+                       return 'l';
+               case 80:
+                       return 't';
                }
-       }
-       else {
-               /* destination address mode nodes don't have outputs */
-               if (is_ia32_irn(irn) && get_ia32_op_type(irn) == ia32_AddrModeD) {
-                       return "MEM";
+       } else {
+               assert(mode_is_int(mode) || mode_is_reference(mode));
+               switch(get_mode_size_bits(mode)) {
+               case 64:
+                       return 'q';
+               case 32:
+                       return 'l';
+               case 16:
+                       return 'w';
+               case 8:
+                       return 'b';
                }
+       }
+       panic("Can't output mode_suffix for %+F\n", mode);
+}
 
-               reg = get_out_reg(irn, pos);
-               if (reg->reg_class == &ia32_reg_classes[CLASS_ia32_vfp])
-                       reg = get_ia32_attr(irn)->x87[pos + 2];
+static
+int produces_result(const ir_node *node) {
+       return !(is_ia32_St(node) ||
+               is_ia32_Store8Bit(node) ||
+               is_ia32_CondJmp(node) ||
+               is_ia32_xCondJmp(node) ||
+               is_ia32_CmpSet(node) ||
+               is_ia32_xCmpSet(node) ||
+               is_ia32_SwitchJmp(node));
+}
+
+static
+const char *ia32_get_reg_name_for_mode(ia32_emit_env_t *env, ir_mode *mode,
+                                       const arch_register_t *reg) {
+       switch(get_mode_size_bits(mode)) {
+               case 8:
+                       return ia32_get_mapped_reg_name(env->isa->regs_8bit, reg);
+               case 16:
+                       return ia32_get_mapped_reg_name(env->isa->regs_16bit, reg);
+               default:
+                       return (char *)arch_register_get_name(reg);
        }
-       return arch_register_get_name(reg);
 }
 
 /**
- * Get the register name for a node.
+ * Add a number to a prefix. This number will not be used a second time.
  */
-static int ia32_get_reg_name(lc_appendable_t *app,
-    const lc_arg_occ_t *occ, const lc_arg_value_t *arg)
+static
+char *get_unique_label(char *buf, size_t buflen, const char *prefix) {
+       static unsigned long id = 0;
+       snprintf(buf, buflen, "%s%lu", prefix, ++id);
+       return buf;
+}
+
+/*************************************************************
+ *             _       _    __   _          _
+ *            (_)     | |  / _| | |        | |
+ *  _ __  _ __ _ _ __ | |_| |_  | |__   ___| |_ __   ___ _ __
+ * | '_ \| '__| | '_ \| __|  _| | '_ \ / _ \ | '_ \ / _ \ '__|
+ * | |_) | |  | | | | | |_| |   | | | |  __/ | |_) |  __/ |
+ * | .__/|_|  |_|_| |_|\__|_|   |_| |_|\___|_| .__/ \___|_|
+ * | |                                       | |
+ * |_|                                       |_|
+ *************************************************************/
+
+// we have no C++ and can't define an implicit ia32_emit_env_t* cast to
+// be_emit_env_t* so we cheat a bit...
+#define be_emit_char(env,c)             be_emit_char(env->emit,c)
+#define be_emit_string(env,s)           be_emit_string(env->emit,s)
+#undef be_emit_cstring
+#define be_emit_cstring(env,x)          { be_emit_string_len(env->emit, x, sizeof(x)-1); }
+#define be_emit_ident(env,i)            be_emit_ident(env->emit,i)
+#define be_emit_write_line(env)         be_emit_write_line(env->emit)
+#define be_emit_finish_line_gas(env,n)  be_emit_finish_line_gas(env->emit,n)
+#define be_emit_pad_comment(env)        be_emit_pad_comment(env->emit)
+
+void ia32_emit_source_register(ia32_emit_env_t *env, const ir_node *node, int pos)
 {
-       const char *buf;
-       ir_node    *irn = arg->v_ptr;
-       int         nr = occ->width - 1;
+       const arch_register_t *reg = get_in_reg(env, node, pos);
+       const char *reg_name = arch_register_get_name(reg);
 
-       if (! irn)
-               return lc_appendable_snadd(app, "(null)", 6);
+       assert(pos < get_irn_arity(node));
 
-       buf = get_ia32_reg_name(irn, nr, occ->conversion == 'S' ? IN_REG : OUT_REG);
+       be_emit_char(env, '%');
+       be_emit_string(env, reg_name);
+}
+
+void ia32_emit_dest_register(ia32_emit_env_t *env, const ir_node *node, int pos) {
+       const arch_register_t *reg = get_out_reg(env, node, pos);
+       const char *reg_name = arch_register_get_name(reg);
 
-       /* append the stupid % to register names */
-       lc_appendable_chadd(app, '%');
-       return lc_appendable_snadd(app, buf, strlen(buf));
+       be_emit_char(env, '%');
+       be_emit_string(env, reg_name);
 }
 
-/**
- * Get the x87 register name for a node.
- */
-static int ia32_get_x87_name(lc_appendable_t *app,
-    const lc_arg_occ_t *occ, const lc_arg_value_t *arg)
+void ia32_emit_x87_name(ia32_emit_env_t *env, const ir_node *node, int pos)
 {
-       const char *buf;
-       ir_node     *irn = arg->v_ptr;
-       int         nr = occ->width - 1;
-       ia32_attr_t *attr;
-
-       if (! irn)
-               return lc_appendable_snadd(app, "(null)", 6);
+       ia32_attr_t *attr = get_ia32_attr(node);
 
-       attr = get_ia32_attr(irn);
-       buf = attr->x87[nr]->name;
-       lc_appendable_chadd(app, '%');
-       return lc_appendable_snadd(app, buf, strlen(buf));
+       assert(pos < 3);
+       be_emit_char(env, '%');
+       be_emit_string(env, attr->x87[pos]->name);
 }
 
-/**
- * Returns the tarval, offset or scale of an ia32 as a string.
- */
-static int ia32_const_to_str(lc_appendable_t *app,
-    const lc_arg_occ_t *occ, const lc_arg_value_t *arg)
+void ia32_emit_immediate(ia32_emit_env_t *env, const ir_node *node)
 {
-       const char *buf;
-       ir_node    *irn = arg->v_ptr;
+       tarval *tv;
+       ir_entity *ent;
+       ident *id;
 
-       if (! irn)
-               return lc_arg_append(app, occ, "(null)", 6);
+       be_emit_char(env, '$');
 
-       if (occ->conversion == 'C') {
-               buf = get_ia32_cnst(irn);
-       }
-       else { /* 'O' */
-               buf = get_ia32_am_offs(irn);
+       switch(get_ia32_immop_type(node)) {
+       case ia32_ImmConst:
+               tv = get_ia32_Immop_tarval(node);
+               id = get_ident_for_tv(tv);
+               break;
+       case ia32_ImmSymConst:
+               ent = get_ia32_Immop_symconst(node);
+               mark_entity_visited(ent);
+               id = get_entity_ld_ident(ent);
+               break;
+       default:
+               assert(0);
+               be_emit_string(env, "BAD");
+               return;
        }
 
-       return buf ? lc_appendable_snadd(app, buf, strlen(buf)) : 0;
+       be_emit_ident(env, id);
 }
 
-/**
- * Determines the SSE suffix depending on the mode.
- */
-static int ia32_get_mode_suffix(lc_appendable_t *app,
-    const lc_arg_occ_t *occ, const lc_arg_value_t *arg)
+void ia32_emit_mode_suffix(ia32_emit_env_t *env, const ir_mode *mode)
+{
+       be_emit_char(env, get_mode_suffix(mode));
+}
+
+void ia32_emit_x87_mode_suffix(ia32_emit_env_t *env, const ir_node *node)
 {
-       ir_node *irn  = arg->v_ptr;
-       ir_mode *mode = get_irn_mode(irn);
+       ir_mode *mode = get_ia32_ls_mode(node);
+       if(mode != NULL)
+               ia32_emit_mode_suffix(env, mode);
+}
 
-       if (mode == mode_T) {
-               mode = (is_ia32_Ld(irn) || is_ia32_St(irn)) ? get_ia32_ls_mode(irn) : get_ia32_res_mode(irn);
+static
+char get_xmm_mode_suffix(ir_mode *mode)
+{
+       assert(mode_is_float(mode));
+       switch(get_mode_size_bits(mode)) {
+       case 32:
+               return 's';
+       case 64:
+               return 'd';
+       default:
+               assert(0);
        }
+       return '%';
+}
 
-       if (! irn)
-               return lc_arg_append(app, occ, "(null)", 6);
+void ia32_emit_xmm_mode_suffix(ia32_emit_env_t *env, const ir_node *node)
+{
+       ir_mode *mode = get_ia32_ls_mode(node);
+       assert(mode != NULL);
+       be_emit_char(env, 's');
+       be_emit_char(env, get_xmm_mode_suffix(mode));
+}
 
-       if (mode_is_float(mode)) {
-               return lc_appendable_chadd(app, get_mode_size_bits(mode) == 32 ? 's' : 'd');
-       }
-       else {
-               return lc_appendable_chadd(app, mode_is_signed(mode) ? 's' : 'z');
-       }
+void ia32_emit_xmm_mode_suffix_s(ia32_emit_env_t *env, const ir_node *node)
+{
+       ir_mode *mode = get_ia32_ls_mode(node);
+       assert(mode != NULL);
+       be_emit_char(env, get_xmm_mode_suffix(mode));
 }
 
-/**
- * Return the ia32 printf arg environment.
- * We use the firm environment with some additional handlers.
- */
-const lc_arg_env_t *ia32_get_arg_env(void) {
-       static lc_arg_env_t *env = NULL;
-
-       static const lc_arg_handler_t ia32_reg_handler   = { ia32_get_arg_type, ia32_get_reg_name };
-       static const lc_arg_handler_t ia32_const_handler = { ia32_get_arg_type, ia32_const_to_str };
-       static const lc_arg_handler_t ia32_mode_handler  = { ia32_get_arg_type, ia32_get_mode_suffix };
-       static const lc_arg_handler_t ia32_x87_handler   = { ia32_get_arg_type, ia32_get_x87_name };
-
-       if(env == NULL) {
-               /* extend the firm printer */
-               env = firm_get_arg_env();
-
-               lc_arg_register(env, "ia32:sreg", 'S', &ia32_reg_handler);
-               lc_arg_register(env, "ia32:dreg", 'D', &ia32_reg_handler);
-               lc_arg_register(env, "ia32:cnst", 'C', &ia32_const_handler);
-               lc_arg_register(env, "ia32:offs", 'O', &ia32_const_handler);
-               lc_arg_register(env, "ia32:mode", 'M', &ia32_mode_handler);
-               lc_arg_register(env, "ia32:x87",  'X', &ia32_x87_handler);
+void ia32_emit_extend_suffix(ia32_emit_env_t *env, const ir_mode *mode)
+{
+       if(get_mode_size_bits(mode) == 32)
+               return;
+       if(mode_is_signed(mode)) {
+               be_emit_char(env, 's');
+       } else {
+               be_emit_char(env, 'z');
        }
+}
 
-       return env;
+static
+void ia32_emit_function_object(ia32_emit_env_t *env, const char *name)
+{
+       switch (be_gas_flavour) {
+       case GAS_FLAVOUR_NORMAL:
+               be_emit_cstring(env, "\t.type\t");
+               be_emit_string(env, name);
+               be_emit_cstring(env, ", @function\n");
+               be_emit_write_line(env);
+               break;
+       case GAS_FLAVOUR_MINGW:
+               be_emit_cstring(env, "\t.def\t");
+               be_emit_string(env, name);
+               be_emit_cstring(env, ";\t.scl\t2;\t.type\t32;\t.endef\n");
+               be_emit_write_line(env);
+               break;
+       default:
+               break;
+       }
 }
 
-static const char *ia32_get_reg_name_for_mode(ia32_emit_env_t *env, ir_mode *mode, const arch_register_t *reg) {
-       switch(get_mode_size_bits(mode)) {
-               case 8:
-                       return ia32_get_mapped_reg_name(env->isa->regs_8bit, reg);
-               case 16:
-                       return ia32_get_mapped_reg_name(env->isa->regs_16bit, reg);
-               default:
-                       return (char *)arch_register_get_name(reg);
+static
+void ia32_emit_function_size(ia32_emit_env_t *env, const char *name)
+{
+       switch (be_gas_flavour) {
+       case GAS_FLAVOUR_NORMAL:
+               be_emit_cstring(env, "\t.size\t");
+               be_emit_string(env, name);
+               be_emit_cstring(env, ", .-");
+               be_emit_string(env, name);
+               be_emit_char(env, '\n');
+               be_emit_write_line(env);
+               break;
+       default:
+               break;
        }
 }
 
+
+
 /**
  * Emits registers and/or address mode of a binary operation.
  */
-const char *ia32_emit_binop(const ir_node *n, ia32_emit_env_t *env) {
-       static char *buf = NULL;
-
-       /* verify that this function is never called on non-AM supporting operations */
-       //assert(get_ia32_am_support(n) != ia32_am_None && "emit binop expects addressmode support");
-
-#define PRODUCES_RESULT(n)   \
-       (!(is_ia32_St(n)      || \
-       is_ia32_Store8Bit(n)  || \
-       is_ia32_CondJmp(n)    || \
-       is_ia32_xCondJmp(n)   || \
-       is_ia32_CmpSet(n)     || \
-       is_ia32_xCmpSet(n)    || \
-       is_ia32_SwitchJmp(n)))
-
-       if (! buf) {
-               buf = xcalloc(1, SNPRINTF_BUF_LEN);
-       }
-       else {
-               memset(buf, 0, SNPRINTF_BUF_LEN);
-       }
-
-       switch(get_ia32_op_type(n)) {
+void ia32_emit_binop(ia32_emit_env_t *env, const ir_node *node) {
+       switch(get_ia32_op_type(node)) {
                case ia32_Normal:
-                       if (is_ia32_ImmConst(n) || is_ia32_ImmSymConst(n)) {
-                               lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%3S, %s", n, get_ia32_cnst(n));
-                       }
-                       else {
-                               const arch_register_t *in1 = get_in_reg(n, 2);
-                               const arch_register_t *in2 = get_in_reg(n, 3);
-                               const arch_register_t *out = PRODUCES_RESULT(n) ? get_out_reg(n, 0) : NULL;
+                       if (is_ia32_ImmConst(node) || is_ia32_ImmSymConst(node)) {
+                               ia32_emit_immediate(env, node);
+                               be_emit_cstring(env, ", ");
+                               ia32_emit_source_register(env, node, 2);
+                       } else {
+                               const arch_register_t *in1 = get_in_reg(env, node, 2);
+                               const arch_register_t *in2 = get_in_reg(env, node, 3);
+                               const arch_register_t *out = produces_result(node) ? get_out_reg(env, node, 0) : NULL;
                                const arch_register_t *in;
                                const char            *in_name;
 
@@ -386,297 +389,182 @@ const char *ia32_emit_binop(const ir_node *n, ia32_emit_env_t *env) {
                                out     = out ? out : in1;
                                in_name = arch_register_get_name(in);
 
-                               if (is_ia32_emit_cl(n)) {
+                               if (is_ia32_emit_cl(node)) {
                                        assert(REGS_ARE_EQUAL(&ia32_gp_regs[REG_ECX], in) && "shift operation needs ecx");
                                        in_name = "cl";
                                }
 
-                               snprintf(buf, SNPRINTF_BUF_LEN, "%%%s, %%%s", arch_register_get_name(out), in_name);
+                               be_emit_char(env, '%');
+                               be_emit_string(env, in_name);
+                               be_emit_cstring(env, ", %");
+                               be_emit_string(env, arch_register_get_name(out));
                        }
                        break;
                case ia32_AddrModeS:
-                       if (is_ia32_ImmConst(n) || is_ia32_ImmSymConst(n)) {
-                               assert(! PRODUCES_RESULT(n) && "Source AM with Const must not produce result");
-                               snprintf(buf, SNPRINTF_BUF_LEN, "%s, %s", get_ia32_cnst(n), ia32_emit_am(n, env));
-                       }
-                       else {
-                               if (PRODUCES_RESULT(n)) {
-                                       lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%1D, %s", n, ia32_emit_am(n, env));
-                               }
-                               else {
-                                       lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%3S, %s", n, ia32_emit_am(n, env));
-                               }
+                       ia32_emit_am(env, node);
+                       be_emit_cstring(env, ", ");
+                       if (is_ia32_ImmConst(node) || is_ia32_ImmSymConst(node)) {
+                               assert(!produces_result(node) && "Source AM with Const must not produce result");
+                               ia32_emit_immediate(env, node);
+                       } else if (produces_result(node)) {
+                               ia32_emit_dest_register(env, node, 0);
+                       } else {
+                               ia32_emit_source_register(env, node, 2);
                        }
                        break;
                case ia32_AddrModeD:
-                       if (is_ia32_ImmConst(n) || is_ia32_ImmSymConst(n)) {
-                               lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%s,%s%s",
-                                       ia32_emit_am(n, env),
-                                       is_ia32_ImmSymConst(n) ? " OFFSET FLAT:" : " ",  /* In case of a symconst we must add OFFSET to */
-                                       get_ia32_cnst(n));                               /* tell the assembler to store it's address.   */
-                       }
-                       else {
-                               const arch_register_t *in1 = get_in_reg(n, 2);
-                               ir_mode              *mode = get_ia32_res_mode(n);
-                               const char           *in_name;
+                       if (is_ia32_ImmConst(node) || is_ia32_ImmSymConst(node)) {
+                               ia32_emit_immediate(env, node);
+                               be_emit_cstring(env, ", ");
+                               ia32_emit_am(env, node);
+                       } else {
+                               const arch_register_t *in1 = get_in_reg(env, node,
+                                                                       get_irn_arity(node) == 5 ? 3 : 2);
+                               ir_mode               *mode = get_ia32_ls_mode(node);
+                               const char            *in_name;
 
-                               mode    = mode ? mode : get_ia32_ls_mode(n);
                                in_name = ia32_get_reg_name_for_mode(env, mode, in1);
 
-                               if (is_ia32_emit_cl(n)) {
+                               if (is_ia32_emit_cl(node)) {
                                        assert(REGS_ARE_EQUAL(&ia32_gp_regs[REG_ECX], in1) && "shift operation needs ecx");
                                        in_name = "cl";
                                }
 
-                               lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%s, %%%s", ia32_emit_am(n, env), in_name);
+                               be_emit_char(env, '%');
+                               be_emit_string(env, in_name);
+                               be_emit_cstring(env, ", ");
+                               ia32_emit_am(env, node);
                        }
                        break;
                default:
                        assert(0 && "unsupported op type");
        }
-
-#undef PRODUCES_RESULT
-
-       return buf;
-}
-
-/**
- * Returns the xxx PTR string for a given mode
- *
- * @param mode      the mode
- * @param x87_insn  if non-zero returns the string for a x87 instruction
- *                  else for a SSE instruction
- */
-static const char *pointer_size(ir_mode *mode, int x87_insn)
-{
-       if (mode) {
-               switch (get_mode_size_bits(mode)) {
-               case 8:  return "BYTE PTR";
-               case 16: return "WORD PTR";
-               case 32: return "DWORD PTR";
-               case 64:
-                       if (x87_insn)
-                               return "QWORD PTR";
-                       return NULL;
-               case 80:
-               case 96: return "XWORD PTR";
-               default: return NULL;
-               }
-       }
-       return NULL;
 }
 
 /**
  * Emits registers and/or address mode of a binary operation.
  */
-const char *ia32_emit_x87_binop(const ir_node *n, ia32_emit_env_t *env) {
-       static char *buf = NULL;
-
-       /* verify that this function is never called on non-AM supporting operations */
-       //assert(get_ia32_am_support(n) != ia32_am_None && "emit binop expects addressmode support");
-
-       if (! buf) {
-               buf = xcalloc(1, SNPRINTF_BUF_LEN);
-       }
-       else {
-               memset(buf, 0, SNPRINTF_BUF_LEN);
-       }
-
-       switch(get_ia32_op_type(n)) {
+void ia32_emit_x87_binop(ia32_emit_env_t *env, const ir_node *node) {
+       switch(get_ia32_op_type(node)) {
                case ia32_Normal:
-                       if (is_ia32_ImmConst(n) || is_ia32_ImmSymConst(n)) {
-                               ir_mode *mode = get_ia32_ls_mode(n);
-                               const char *p = pointer_size(mode, 1);
-                               lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%s %s", p, get_ia32_cnst(n));
-                       }
-                       else {
-                               ia32_attr_t *attr = get_ia32_attr(n);
+                       if (is_ia32_ImmConst(node) || is_ia32_ImmSymConst(node)) {
+                               // should not happen...
+                               assert(0);
+                       } else {
+                               ia32_attr_t *attr = get_ia32_attr(node);
                                const arch_register_t *in1 = attr->x87[0];
                                const arch_register_t *in2 = attr->x87[1];
                                const arch_register_t *out = attr->x87[2];
                                const arch_register_t *in;
-                               const char            *in_name;
 
-                               in      = out ? (REGS_ARE_EQUAL(out, in2) ? in1 : in2) : in2;
-                               out     = out ? out : in1;
-                               in_name = arch_register_get_name(in);
+                               in  = out ? (REGS_ARE_EQUAL(out, in2) ? in1 : in2) : in2;
+                               out = out ? out : in1;
 
-                               snprintf(buf, SNPRINTF_BUF_LEN, "%%%s, %%%s", arch_register_get_name(out), in_name);
+                               be_emit_char(env, '%');
+                               be_emit_string(env, arch_register_get_name(in));
+                               be_emit_cstring(env, ", %");
+                               be_emit_string(env, arch_register_get_name(out));
                        }
                        break;
                case ia32_AddrModeS:
                case ia32_AddrModeD:
-                       lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%s", ia32_emit_am(n, env));
+                       ia32_emit_am(env, node);
                        break;
                default:
                        assert(0 && "unsupported op type");
        }
-
-       return buf;
 }
 
 /**
  * Emits registers and/or address mode of a unary operation.
  */
-const char *ia32_emit_unop(const ir_node *n, ia32_emit_env_t *env) {
-       static char *buf = NULL;
-
-       if (! buf) {
-               buf = xcalloc(1, SNPRINTF_BUF_LEN);
-       }
-       else {
-               memset(buf, 0, SNPRINTF_BUF_LEN);
-       }
-
-       switch(get_ia32_op_type(n)) {
+void ia32_emit_unop(ia32_emit_env_t *env, const ir_node *node) {
+       switch(get_ia32_op_type(node)) {
                case ia32_Normal:
-                       if (is_ia32_ImmConst(n) || is_ia32_ImmSymConst(n)) {
-                               lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%C", n);
-                       }
-                       else {
-                               if (is_ia32_MulS(n) || is_ia32_Mulh(n)) {
-                                       /* MulS and Mulh implicitly multiply by EAX */
-                                       lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%4S", n);
+                       if (is_ia32_ImmConst(node) || is_ia32_ImmSymConst(node)) {
+                               ia32_emit_immediate(env, node);
+                       } else {
+                               if (is_ia32_Mul(node) || is_ia32_IMul1OP(node)) {
+                                       ia32_emit_source_register(env, node, 3);
+                               } else if(is_ia32_IDiv(node) || is_ia32_Div(node)) {
+                                       ia32_emit_source_register(env, node, 4);
+                               } else if(is_ia32_Push(node)) {
+                                       ia32_emit_source_register(env, node, 2);
+                               } else if(is_ia32_Pop(node)) {
+                                       ia32_emit_dest_register(env, node, 1);
+                               } else {
+                                       ia32_emit_dest_register(env, node, 0);
                                }
-                               else
-                                       lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%1D", n);
                        }
                        break;
+               case ia32_AddrModeS:
                case ia32_AddrModeD:
-                       snprintf(buf, SNPRINTF_BUF_LEN, "%s", ia32_emit_am(n, env));
+                       ia32_emit_am(env, node);
                        break;
                default:
                        assert(0 && "unsupported op type");
        }
-
-       return buf;
 }
 
 /**
  * Emits address mode.
  */
-const char *ia32_emit_am(const ir_node *n, ia32_emit_env_t *env) {
-       ia32_am_flavour_t am_flav    = get_ia32_am_flavour(n);
-       int               had_output = 0;
-       char              *s;
-       const char        *p;
-       int               size;
-       static struct obstack *obst  = NULL;
-       ir_mode *mode = get_ia32_ls_mode(n);
-
-       if (! is_ia32_Lea(n))
-               assert(mode && "AM node must have ls_mode attribute set.");
-
-       if (! obst) {
-               obst = xcalloc(1, sizeof(*obst));
-       }
-       else {
-               obstack_free(obst, NULL);
-       }
-
-       /* obstack_free with NULL results in an uninitialized obstack */
-       obstack_init(obst);
-
-       p = pointer_size(mode, has_x87_register(n));
-       if (p)
-               obstack_printf(obst, "%s ", p);
-
-       /* emit address mode symconst */
-       if (get_ia32_am_sc(n)) {
-               if (is_ia32_am_sc_sign(n))
-                       obstack_printf(obst, "-");
-               obstack_printf(obst, "%s", get_id_str(get_ia32_am_sc(n)));
-       }
-
-       if (am_flav & ia32_B) {
-               obstack_printf(obst, "[");
-               lc_eoprintf(ia32_get_arg_env(), obst, "%1S", n);
-               had_output = 1;
+void ia32_emit_am(ia32_emit_env_t *env, const ir_node *node) {
+       ia32_am_flavour_t am_flav = get_ia32_am_flavour(node);
+       ir_entity *ent = get_ia32_am_sc(node);
+       int offs = get_ia32_am_offs_int(node);
+
+       /* just to be sure... */
+       assert(!is_ia32_use_frame(node) || get_ia32_frame_ent(node) != NULL);
+
+       /* emit offset */
+       if (ent != NULL) {
+               ident *id;
+
+               mark_entity_visited(ent);
+               id = get_entity_ld_ident(ent);
+               if (is_ia32_am_sc_sign(node))
+                       be_emit_char(env, '-');
+               be_emit_ident(env, id);
+
+               if(get_entity_owner(ent) == get_tls_type()) {
+                       if (get_entity_visibility(ent) == visibility_external_allocated) {
+                               be_emit_cstring(env, "@INDNTPOFF");
+                       } else {
+                               be_emit_cstring(env, "@NTPOFF");
+                       }
+               }
        }
 
-       if (am_flav & ia32_I) {
-               if (had_output) {
-                       obstack_printf(obst, "+");
-               }
-               else {
-                       obstack_printf(obst, "[");
+       if(offs != 0) {
+               if(ent != NULL) {
+                       be_emit_irprintf(env->emit, "%+d", offs);
+               } else {
+                       be_emit_irprintf(env->emit, "%d", offs);
                }
+       }
 
-               lc_eoprintf(ia32_get_arg_env(), obst, "%2S", n);
+       if (am_flav & (ia32_B | ia32_I)) {
+               be_emit_char(env, '(');
 
-               if (am_flav & ia32_S) {
-                       obstack_printf(obst, "*%d", 1 << get_ia32_am_scale(n));
+               /* emit base */
+               if (am_flav & ia32_B) {
+                       ia32_emit_source_register(env, node, 0);
                }
 
-               had_output = 1;
-       }
-
-       if (am_flav & ia32_O) {
-               s = get_ia32_am_offs(n);
+               /* emit index + scale */
+               if (am_flav & ia32_I) {
+                       be_emit_char(env, ',');
+                       ia32_emit_source_register(env, node, 1);
 
-               if (s) {
-                       /* omit explicit + if there was no base or index */
-                       if (! had_output) {
-                               obstack_printf(obst, "[");
-                               if (s[0] == '+')
-                                       s++;
+                       if (am_flav & ia32_S) {
+                               be_emit_irprintf(env->emit, ",%d", 1 << get_ia32_am_scale(node));
                        }
-
-                       obstack_printf(obst, s);
-                       had_output = 1;
                }
+               be_emit_char(env, ')');
        }
-
-       if (had_output)
-               obstack_printf(obst, "] ");
-
-       size        = obstack_object_size(obst);
-       s           = obstack_finish(obst);
-       s[size - 1] = '\0';
-
-       return s;
-}
-
-/**
- * emit an address
- */
-const char *ia32_emit_adr(const ir_node *irn, ia32_emit_env_t *env)
-{
-       static char buf[SNPRINTF_BUF_LEN];
-       ir_mode    *mode = get_ia32_ls_mode(irn);
-       const char *adr  = get_ia32_cnst(irn);
-       const char *pref = pointer_size(mode, has_x87_register(irn));
-
-       snprintf(buf, SNPRINTF_BUF_LEN, "%s %s", pref ? pref : "", adr);
-       return buf;
-}
-
-/**
- * Formated print of commands and comments.
- */
-static void ia32_fprintf_format(FILE *F, const ir_node *irn, char *cmd_buf, char *cmnt_buf) {
-       unsigned lineno;
-       const char *name = irn ? be_retrieve_dbg_info(get_irn_dbg_info((ir_node *)irn), &lineno) : NULL;
-
-       if (name)
-               fprintf(F, "\t%-35s %-60s /* %s:%u */\n", cmd_buf, cmnt_buf, name, lineno);
-       else
-               fprintf(F, "\t%-35s %-60s\n", cmd_buf, cmnt_buf);
 }
 
-
-
-/**
- * Add a number to a prefix. This number will not be used a second time.
- */
-static char *get_unique_label(char *buf, size_t buflen, const char *prefix) {
-       static unsigned long id = 0;
-       snprintf(buf, buflen, "%s%lu", prefix, ++id);
-       return buf;
-}
-
-
-
 /*************************************************
  *                 _ _                         _
  *                (_) |                       | |
@@ -701,29 +589,31 @@ struct cmp2conditon_t {
 /*
  * positive conditions for signed compares
  */
-static const struct cmp2conditon_t cmp2condition_s[] = {
-  { NULL,              pn_Cmp_False },  /* always false */
-  { "e",               pn_Cmp_Eq },     /* == */
-  { "l",               pn_Cmp_Lt },     /* < */
-  { "le",              pn_Cmp_Le },     /* <= */
-  { "g",               pn_Cmp_Gt },     /* > */
-  { "ge",              pn_Cmp_Ge },     /* >= */
-  { "ne",              pn_Cmp_Lg },     /* != */
-  { "ordered",         pn_Cmp_Leg },    /* Floating point: ordered */
-  { "unordered",       pn_Cmp_Uo },     /* FLoting point: unordered */
-  { "unordered or ==", pn_Cmp_Ue },     /* Floating point: unordered or == */
-  { "unordered or <",  pn_Cmp_Ul },     /* Floating point: unordered or < */
-  { "unordered or <=", pn_Cmp_Ule },    /* Floating point: unordered or <= */
-  { "unordered or >",  pn_Cmp_Ug },     /* Floating point: unordered or > */
-  { "unordered or >=", pn_Cmp_Uge },    /* Floating point: unordered or >= */
-  { "unordered or !=", pn_Cmp_Ne },     /* Floating point: unordered or != */
-  { NULL,              pn_Cmp_True },   /* always true */
+static
+const struct cmp2conditon_t cmp2condition_s[] = {
+       { NULL,              pn_Cmp_False },  /* always false */
+       { "e",               pn_Cmp_Eq },     /* == */
+       { "l",               pn_Cmp_Lt },     /* < */
+       { "le",              pn_Cmp_Le },     /* <= */
+       { "g",               pn_Cmp_Gt },     /* > */
+       { "ge",              pn_Cmp_Ge },     /* >= */
+       { "ne",              pn_Cmp_Lg },     /* != */
+       { NULL,              pn_Cmp_Leg},     /* Floating point: ordered */
+       { NULL,              pn_Cmp_Uo },     /* Floating point: unordered */
+       { "e",               pn_Cmp_Ue },     /* Floating point: unordered or == */
+       { "b",               pn_Cmp_Ul },     /* Floating point: unordered or < */
+       { "be",              pn_Cmp_Ule },    /* Floating point: unordered or <= */
+       { "a",               pn_Cmp_Ug },     /* Floating point: unordered or > */
+       { "ae",              pn_Cmp_Uge },    /* Floating point: unordered or >= */
+       { "ne",              pn_Cmp_Ne },     /* Floating point: unordered or != */
+       { NULL,              pn_Cmp_True },   /* always true */
 };
 
 /*
  * positive conditions for unsigned compares
  */
-static const struct cmp2conditon_t cmp2condition_u[] = {
+static
+const struct cmp2conditon_t cmp2condition_u[] = {
        { NULL,              pn_Cmp_False },  /* always false */
        { "e",               pn_Cmp_Eq },     /* == */
        { "b",               pn_Cmp_Lt },     /* < */
@@ -731,43 +621,47 @@ static const struct cmp2conditon_t cmp2condition_u[] = {
        { "a",               pn_Cmp_Gt },     /* > */
        { "ae",              pn_Cmp_Ge },     /* >= */
        { "ne",              pn_Cmp_Lg },     /* != */
-       { "ordered",         pn_Cmp_Leg },    /* Floating point: ordered */
-       { "unordered",       pn_Cmp_Uo },     /* FLoting point: unordered */
-       { "unordered or ==", pn_Cmp_Ue },     /* Floating point: unordered or == */
-       { "unordered or <",  pn_Cmp_Ul },     /* Floating point: unordered or < */
-       { "unordered or <=", pn_Cmp_Ule },    /* Floating point: unordered or <= */
-       { "unordered or >",  pn_Cmp_Ug },     /* Floating point: unordered or > */
-       { "unordered or >=", pn_Cmp_Uge },    /* Floating point: unordered or >= */
-       { "unordered or !=", pn_Cmp_Ne },     /* Floating point: unordered or != */
        { NULL,              pn_Cmp_True },   /* always true */
 };
 
 /*
  * returns the condition code
  */
-static const char *get_cmp_suffix(int cmp_code, int unsigned_cmp)
+static
+const char *get_cmp_suffix(int cmp_code)
 {
-       assert(cmp2condition_s[cmp_code].num == cmp_code);
-       assert(cmp2condition_u[cmp_code].num == cmp_code);
+       assert( (cmp2condition_s[cmp_code & 15].num) == (cmp_code & 15));
+       assert( (cmp2condition_u[cmp_code & 7].num) == (cmp_code & 7));
+
+       if((cmp_code & ia32_pn_Cmp_Unsigned)) {
+               return cmp2condition_u[cmp_code & 7].name;
+       } else {
+               return cmp2condition_s[cmp_code & 15].name;
+       }
+}
 
-       return unsigned_cmp ? cmp2condition_u[cmp_code & 7].name : cmp2condition_s[cmp_code & 7].name;
+void ia32_emit_cmp_suffix(ia32_emit_env_t *env, long pnc)
+{
+       be_emit_string(env, get_cmp_suffix(pnc));
 }
 
+
 /**
  * Returns the target block for a control flow node.
  */
-static ir_node *get_cfop_target_block(const ir_node *irn) {
+static
+ir_node *get_cfop_target_block(const ir_node *irn) {
        return get_irn_link(irn);
 }
 
 /**
  * Returns the target label for a control flow node.
  */
-static char *get_cfop_target(const ir_node *irn, char *buf) {
-       ir_node *bl = get_cfop_target_block(irn);
+void ia32_emit_cfop_target(ia32_emit_env_t * env, const ir_node *node) {
+       ir_node *block = get_cfop_target_block(node);
 
-       snprintf(buf, SNPRINTF_BUF_LEN, BLOCK_PREFIX("%ld"), get_irn_node_nr(bl));
-       return buf;
+       be_emit_cstring(env, BLOCK_PREFIX);
+       be_emit_irprintf(env->emit, "%d", get_irn_node_nr(block));
 }
 
 /** Return the next block in Block schedule */
@@ -778,13 +672,14 @@ static ir_node *next_blk_sched(const ir_node *block) {
 /**
  * Returns the Proj with projection number proj and NOT mode_M
  */
-static ir_node *get_proj(const ir_node *irn, long proj) {
+static
+ir_node *get_proj(const ir_node *node, long proj) {
        const ir_edge_t *edge;
        ir_node         *src;
 
-       assert(get_irn_mode(irn) == mode_T && "expected mode_T node");
+       assert(get_irn_mode(node) == mode_T && "expected mode_T node");
 
-       foreach_out_edge(irn, edge) {
+       foreach_out_edge(node, edge) {
                src = get_edge_src_irn(edge);
 
                assert(is_Proj(src) && "Proj expected");
@@ -800,369 +695,416 @@ static ir_node *get_proj(const ir_node *irn, long proj) {
 /**
  * Emits the jump sequence for a conditional jump (cmp + jmp_true + jmp_false)
  */
-static void finish_CondJmp(FILE *F, const ir_node *irn, ir_mode *mode) {
-       const ir_node   *proj1, *proj2 = NULL;
-       const ir_node   *block, *next_bl = NULL;
-       char buf[SNPRINTF_BUF_LEN];
-       char cmd_buf[SNPRINTF_BUF_LEN];
-       char cmnt_buf[SNPRINTF_BUF_LEN];
+static
+void finish_CondJmp(ia32_emit_env_t *env, const ir_node *node, ir_mode *mode,
+                    long pnc) {
+       const ir_node *proj_true;
+       const ir_node *proj_false;
+       const ir_node *block;
+       const ir_node *next_block;
+       int flipped = 0;
 
        /* get both Proj's */
-       proj1 = get_proj(irn, pn_Cond_true);
-       assert(proj1 && "CondJmp without true Proj");
+       proj_true = get_proj(node, pn_Cond_true);
+       assert(proj_true && "CondJmp without true Proj");
 
-       proj2 = get_proj(irn, pn_Cond_false);
-       assert(proj2 && "CondJmp without false Proj");
+       proj_false = get_proj(node, pn_Cond_false);
+       assert(proj_false && "CondJmp without false Proj");
 
        /* for now, the code works for scheduled and non-schedules blocks */
-       block = get_nodes_block(irn);
+       block = get_nodes_block(node);
 
        /* we have a block schedule */
-       next_bl = next_blk_sched(block);
+       next_block = next_blk_sched(block);
 
-       if (get_cfop_target_block(proj1) == next_bl) {
+       if (get_cfop_target_block(proj_true) == next_block) {
                /* exchange both proj's so the second one can be omitted */
-               const ir_node *t = proj1;
-               proj1 = proj2;
-               proj2 = t;
-       }
+               const ir_node *t = proj_true;
 
-       /* the first Proj must always be created */
-       if (get_Proj_proj(proj1) == pn_Cond_true) {
-               snprintf(cmd_buf, SNPRINTF_BUF_LEN, "j%s %s",
-                                       get_cmp_suffix(get_ia32_pncode(irn), ! mode_is_signed(get_irn_mode(get_irn_n(irn, 0)))),
-                                       get_cfop_target(proj1, buf));
-               snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* cmp(a, b) == TRUE */");
+               proj_true  = proj_false;
+               proj_false = t;
+               flipped    = 1;
+               pnc        = get_negated_pnc(pnc, mode);
        }
-       else  {
-               snprintf(cmd_buf, SNPRINTF_BUF_LEN, "j%s %s",
-                                       get_cmp_suffix(get_negated_pnc(get_ia32_pncode(irn), mode),
-                                       ! mode_is_signed(get_irn_mode(get_irn_n(irn, 0)))),
-                                       get_cfop_target(proj1, buf));
-               snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* cmp(a, b) == FALSE */");
+
+       /* in case of unordered compare, check for parity */
+       if (pnc & pn_Cmp_Uo) {
+               be_emit_cstring(env, "\tjp ");
+               ia32_emit_cfop_target(env, proj_true);
+               be_emit_finish_line_gas(env, proj_true);
        }
-       IA32_DO_EMIT(irn);
+
+       be_emit_cstring(env, "\tj");
+       ia32_emit_cmp_suffix(env, pnc);
+       be_emit_char(env, ' ');
+       ia32_emit_cfop_target(env, proj_true);
+       be_emit_finish_line_gas(env, proj_true);
 
        /* the second Proj might be a fallthrough */
-       if (get_cfop_target_block(proj2) != next_bl) {
-               snprintf(cmd_buf, SNPRINTF_BUF_LEN, "jmp %s", get_cfop_target(proj2, buf));
-               snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* otherwise */");
-       }
-       else {
-               cmd_buf[0] = '\0';
-               snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* fallthrough %s */", get_cfop_target(proj2, buf));
+       if (get_cfop_target_block(proj_false) != next_block) {
+               be_emit_cstring(env, "\tjmp ");
+               ia32_emit_cfop_target(env, proj_false);
+               be_emit_finish_line_gas(env, proj_false);
+       } else {
+               be_emit_cstring(env, "\t/* fallthrough to");
+               ia32_emit_cfop_target(env, proj_false);
+               be_emit_cstring(env, " */");
+               be_emit_finish_line_gas(env, proj_false);
        }
-       IA32_DO_EMIT(irn);
 }
 
 /**
  * Emits code for conditional jump.
  */
-static void CondJmp_emitter(const ir_node *irn, ia32_emit_env_t *env) {
-       FILE *F = env->out;
-       char cmd_buf[SNPRINTF_BUF_LEN];
-       char cmnt_buf[SNPRINTF_BUF_LEN];
+static
+void CondJmp_emitter(ia32_emit_env_t *env, const ir_node *node) {
+       be_emit_cstring(env, "\tcmp ");
+       ia32_emit_binop(env, node);
+       be_emit_finish_line_gas(env, node);
 
-       snprintf(cmd_buf, SNPRINTF_BUF_LEN, "cmp %s", ia32_emit_binop(irn, env));
-       lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F */", irn);
-       IA32_DO_EMIT(irn);
-       finish_CondJmp(F, irn, get_ia32_res_mode(irn));
+       finish_CondJmp(env, node, mode_Iu, get_ia32_pncode(node));
 }
 
 /**
  * Emits code for conditional jump with two variables.
  */
-static void emit_ia32_CondJmp(const ir_node *irn, ia32_emit_env_t *env) {
-       CondJmp_emitter(irn, env);
+static
+void emit_ia32_CondJmp(ia32_emit_env_t *env, const ir_node *node) {
+       CondJmp_emitter(env, node);
 }
 
 /**
  * Emits code for conditional test and jump.
  */
-static void TestJmp_emitter(const ir_node *irn, ia32_emit_env_t *env) {
-
-#define IA32_IS_IMMOP (is_ia32_ImmConst(irn) || is_ia32_ImmSymConst(irn))
-
-       FILE       *F   = env->out;
-       const char *op1 = arch_register_get_name(get_in_reg(irn, 0));
-       const char *op2 = IA32_IS_IMMOP ? get_ia32_cnst(irn) : NULL;
-       char        cmd_buf[SNPRINTF_BUF_LEN];
-       char        cmnt_buf[SNPRINTF_BUF_LEN];
-
-       if (! op2)
-               op2 = arch_register_get_name(get_in_reg(irn, 1));
-
-       snprintf(cmd_buf, SNPRINTF_BUF_LEN, "test %%%s,%s%s ", op1, IA32_IS_IMMOP ? " " : " %", op2);
-       lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F */", irn);
-
-       IA32_DO_EMIT(irn);
-       finish_CondJmp(F, irn, get_ia32_res_mode(irn));
-
-#undef IA32_IS_IMMOP
+static
+void TestJmp_emitter(ia32_emit_env_t *env, const ir_node *node) {
+       if(is_ia32_ImmSymConst(node) || is_ia32_ImmConst(node)) {
+               be_emit_cstring(env, "\ttest ");
+               ia32_emit_immediate(env, node);
+               be_emit_cstring(env, ", ");
+               ia32_emit_source_register(env, node, 0);
+               be_emit_finish_line_gas(env, node);
+       } else {
+               be_emit_cstring(env, "\ttest ");
+               ia32_emit_source_register(env, node, 1);
+               be_emit_cstring(env, ", ");
+               ia32_emit_source_register(env, node, 0);
+               be_emit_finish_line_gas(env, node);
+       }
+       finish_CondJmp(env, node, mode_Iu, get_ia32_pncode(node));
 }
 
 /**
  * Emits code for conditional test and jump with two variables.
  */
-static void emit_ia32_TestJmp(const ir_node *irn, ia32_emit_env_t *env) {
-       TestJmp_emitter(irn, env);
+static
+void emit_ia32_TestJmp(ia32_emit_env_t *env, const ir_node *node) {
+       TestJmp_emitter(env, node);
 }
 
-static void emit_ia32_CJmp(const ir_node *irn, ia32_emit_env_t *env) {
-       FILE *F = env->out;
-       char cmd_buf[SNPRINTF_BUF_LEN];
-       char cmnt_buf[SNPRINTF_BUF_LEN];
+static
+void emit_ia32_CJmp(ia32_emit_env_t *env, const ir_node *node) {
+       be_emit_cstring(env, "/* omitted redundant test */");
+       be_emit_finish_line_gas(env, node);
 
-       snprintf(cmd_buf, SNPRINTF_BUF_LEN, " ");
-       lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F omitted redundant test */", irn);
-       IA32_DO_EMIT(irn);
-       finish_CondJmp(F, irn, get_ia32_res_mode(irn));
+       finish_CondJmp(env, node, mode_Is, get_ia32_pncode(node));
 }
 
-static void emit_ia32_CJmpAM(const ir_node *irn, ia32_emit_env_t *env) {
-       FILE *F = env->out;
-       char cmd_buf[SNPRINTF_BUF_LEN];
-       char cmnt_buf[SNPRINTF_BUF_LEN];
+static
+void emit_ia32_CJmpAM(ia32_emit_env_t *env, const ir_node *node) {
+       be_emit_cstring(env, "/* omitted redundant test/cmp */");
+       be_emit_finish_line_gas(env, node);
 
-       snprintf(cmd_buf, SNPRINTF_BUF_LEN, " ");
-       lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F omitted redundant test/cmp */", irn);
-       IA32_DO_EMIT(irn);
-       finish_CondJmp(F, irn, get_ia32_res_mode(irn));
+       finish_CondJmp(env, node, mode_Is, get_ia32_pncode(node));
 }
 
 /**
  * Emits code for conditional SSE floating point jump with two variables.
  */
-static void emit_ia32_xCondJmp(ir_node *irn, ia32_emit_env_t *env) {
-       FILE *F = env->out;
-       char cmd_buf[SNPRINTF_BUF_LEN];
-       char cmnt_buf[SNPRINTF_BUF_LEN];
-
-       lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "ucomis%M %s", irn, ia32_emit_binop(irn, env));
-       lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F */", irn);
-       IA32_DO_EMIT(irn);
-       finish_CondJmp(F, irn, get_ia32_res_mode(irn));
+static
+void emit_ia32_xCondJmp(ia32_emit_env_t *env, const ir_node *node) {
+       be_emit_cstring(env, "\tucomi");
+       ia32_emit_xmm_mode_suffix(env, node);
+       be_emit_char(env, ' ');
+       ia32_emit_binop(env, node);
+       be_emit_finish_line_gas(env, node);
 
+       finish_CondJmp(env, node, mode_F, get_ia32_pncode(node));
 }
 
 /**
  * Emits code for conditional x87 floating point jump with two variables.
  */
-static void emit_ia32_x87CondJmp(ir_node *irn, ia32_emit_env_t *env) {
-       FILE *F = env->out;
-       char cmd_buf[SNPRINTF_BUF_LEN];
-       char cmnt_buf[SNPRINTF_BUF_LEN];
-       ia32_attr_t *attr = get_ia32_attr(irn);
+static
+void emit_ia32_x87CondJmp(ia32_emit_env_t *env, const ir_node *node) {
+       ia32_attr_t *attr = get_ia32_attr(node);
        const char *reg = attr->x87[1]->name;
-       const char *instr = "fcom";
-       int reverse = 0;
+       long pnc = get_ia32_pncode(node);
 
-       switch (get_ia32_pncode(irn)) {
+       switch (get_ia32_irn_opcode(node)) {
        case iro_ia32_fcomrJmp:
-               reverse = 1;
+               pnc = get_inversed_pnc(pnc);
+               reg = attr->x87[0]->name;
        case iro_ia32_fcomJmp:
        default:
-               instr = "fucom";
+               be_emit_cstring(env, "\tfucom ");
                break;
        case iro_ia32_fcomrpJmp:
-               reverse = 1;
+               pnc = get_inversed_pnc(pnc);
+               reg = attr->x87[0]->name;
        case iro_ia32_fcompJmp:
-               instr = "fucomp";
+               be_emit_cstring(env, "\tfucomp ");
                break;
        case iro_ia32_fcomrppJmp:
-               reverse = 1;
+               pnc = get_inversed_pnc(pnc);
        case iro_ia32_fcomppJmp:
-               instr = "fucompp";
+               be_emit_cstring(env, "\tfucompp ");
                reg = "";
                break;
        }
 
-       if (reverse)
-               set_ia32_pncode(irn, (long)get_negated_pnc(get_ia32_pncode(irn), mode_Is));
+       if(reg[0] != '\0') {
+               be_emit_char(env, '%');
+               be_emit_string(env, reg);
+       }
+       be_emit_finish_line_gas(env, node);
 
-       snprintf(cmd_buf, SNPRINTF_BUF_LEN, "%s %%%s", instr, reg);
-       lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F */", irn);
-       IA32_DO_EMIT(irn);
-       lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "fnstsw %%ax", irn);
-       snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* Store x87 FPU Control Word */");
-       IA32_DO_EMIT(irn);
-       snprintf(cmd_buf, SNPRINTF_BUF_LEN, "sahf");
-       snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* Store ah into flags */");
-       IA32_DO_EMIT(irn);
+       be_emit_cstring(env, "\tfnstsw %ax");
+       be_emit_finish_line_gas(env, node);
+       be_emit_cstring(env, "\tsahf");
+       be_emit_finish_line_gas(env, node);
 
-       finish_CondJmp(F, irn, mode_Is);
+       finish_CondJmp(env, node, mode_E, pnc);
 }
 
-static void CMov_emitter(ir_node *irn, ia32_emit_env_t *env) {
-       FILE               *F       = env->out;
-       const lc_arg_env_t *arg_env = ia32_get_arg_env();
-       const char *cmp_suffix = get_cmp_suffix(get_ia32_pncode(irn), ! mode_is_signed(get_irn_mode(get_irn_n(irn, 0))));
-       int is_PsiCondCMov     = is_ia32_PsiCondCMov(irn);
-
-       char cmd_buf[SNPRINTF_BUF_LEN];
-       char cmnt_buf[SNPRINTF_BUF_LEN];
+static
+void CMov_emitter(ia32_emit_env_t *env, const ir_node *node) {
+       long pnc = get_ia32_pncode(node);
+       int is_PsiCondCMov = is_ia32_PsiCondCMov(node);
+       int idx_left  = 2 - is_PsiCondCMov;
+       int idx_right = 3 - is_PsiCondCMov;
        const arch_register_t *in1, *in2, *out;
 
-       out = arch_get_irn_register(env->arch_env, irn);
-       in1 = arch_get_irn_register(env->arch_env, get_irn_n(irn, 2 - is_PsiCondCMov));
-       in2 = arch_get_irn_register(env->arch_env, get_irn_n(irn, 3 - is_PsiCondCMov));
+       out = arch_get_irn_register(env->arch_env, node);
+       in1 = arch_get_irn_register(env->arch_env, get_irn_n(node, idx_left));
+       in2 = arch_get_irn_register(env->arch_env, get_irn_n(node, idx_right));
 
        /* we have to emit the cmp first, because the destination register */
        /* could be one of the compare registers                           */
-       if (is_ia32_CmpCMov(irn)) {
-               lc_esnprintf(arg_env, cmd_buf, SNPRINTF_BUF_LEN, "cmp %1S, %2S", irn, irn);
-       }
-       else if (is_ia32_xCmpCMov(irn)) {
-               lc_esnprintf(arg_env, cmd_buf, SNPRINTF_BUF_LEN, "ucomis%M %1S, %2S", get_irn_n(irn, 0), irn, irn);
-       }
-       else if (is_PsiCondCMov) {
+       if (is_ia32_CmpCMov(node)) {
+               be_emit_cstring(env, "\tcmp ");
+               ia32_emit_source_register(env, node, 1);
+               be_emit_cstring(env, ", ");
+               ia32_emit_source_register(env, node, 0);
+       } else if (is_ia32_xCmpCMov(node)) {
+               be_emit_cstring(env, "\tucomis");
+               ia32_emit_mode_suffix(env, get_irn_mode(node));
+               be_emit_char(env, ' ');
+               ia32_emit_source_register(env, node, 1);
+               be_emit_cstring(env, ", ");
+               ia32_emit_source_register(env, node, 0);
+       } else if (is_PsiCondCMov) {
                /* omit compare because flags are already set by And/Or */
-               snprintf(cmd_buf, SNPRINTF_BUF_LEN, " ");
-       }
-       else {
+               be_emit_cstring(env, "\ttest ");
+               ia32_emit_source_register(env, node, 0);
+               be_emit_cstring(env, ", ");
+               ia32_emit_source_register(env, node, 0);
+       } else {
                assert(0 && "unsupported CMov");
        }
-       snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* Psi condition */" );
-       IA32_DO_EMIT(irn);
+       be_emit_finish_line_gas(env, node);
 
        if (REGS_ARE_EQUAL(out, in2)) {
                /* best case: default in == out -> do nothing */
-       }
-       else if (REGS_ARE_EQUAL(out, in1)) {
+       } else if (REGS_ARE_EQUAL(out, in1)) {
+               ir_node *n = (ir_node*) node;
                /* true in == out -> need complement compare and exchange true and default in */
-               ir_node *t = get_irn_n(irn, 2);
-               set_irn_n(irn, 2, get_irn_n(irn, 3));
-               set_irn_n(irn, 3, t);
-
-               cmp_suffix = get_cmp_suffix(get_inversed_pnc(get_ia32_pncode(irn)), ! mode_is_signed(get_irn_mode(get_irn_n(irn, 0))));
+               ir_node *t = get_irn_n(n, idx_left);
+               set_irn_n(n, idx_left, get_irn_n(n, idx_right));
+               set_irn_n(n, idx_right, t);
 
-       }
-       else {
+               pnc = get_negated_pnc(pnc, get_irn_mode(node));
+       else {
                /* out is different from in: need copy default -> out */
-               lc_esnprintf(arg_env, cmd_buf, SNPRINTF_BUF_LEN, "mov %1D, %4S", irn, irn);
-               lc_esnprintf(arg_env, cmnt_buf, SNPRINTF_BUF_LEN, "/* copy default -> out */" );
-               IA32_DO_EMIT(irn);
+               if (is_PsiCondCMov) {
+                       be_emit_cstring(env, "\tmovl ");
+                       ia32_emit_dest_register(env, node, 2);
+                       be_emit_cstring(env, ", ");
+                       ia32_emit_dest_register(env, node, 0);
+               } else {
+                       be_emit_cstring(env, "\tmovl ");
+                       ia32_emit_source_register(env, node, 3);
+                       be_emit_cstring(env, ", ");
+                       ia32_emit_dest_register(env, node, 0);
+               }
+               be_emit_finish_line_gas(env, node);
        }
 
-       lc_esnprintf(arg_env, cmd_buf, SNPRINTF_BUF_LEN, "cmov%s %1D, %3S", cmp_suffix, irn, irn);
-       lc_esnprintf(arg_env, cmnt_buf, SNPRINTF_BUF_LEN, "/* condition is true case */" );
-       IA32_DO_EMIT(irn);
+       if (is_PsiCondCMov) {
+               be_emit_cstring(env, "\tcmov");
+               ia32_emit_cmp_suffix(env, pnc);
+               be_emit_cstring(env, "l ");
+               ia32_emit_source_register(env, node, 1);
+               be_emit_cstring(env, ", ");
+               ia32_emit_dest_register(env, node, 0);
+       } else {
+               be_emit_cstring(env, "\tcmov");
+               ia32_emit_cmp_suffix(env, pnc);
+               be_emit_cstring(env, "l ");
+               ia32_emit_source_register(env, node, 2);
+               be_emit_cstring(env, ", ");
+               ia32_emit_dest_register(env, node, 0);
+       }
+       be_emit_finish_line_gas(env, node);
 }
 
-static void emit_ia32_CmpCMov(ir_node *irn, ia32_emit_env_t *env) {
-       CMov_emitter(irn, env);
+static
+void emit_ia32_CmpCMov(ia32_emit_env_t *env, const ir_node *node) {
+       CMov_emitter(env, node);
 }
 
-static void emit_ia32_PsiCondCMov(ir_node *irn, ia32_emit_env_t *env) {
-       CMov_emitter(irn, env);
+static
+void emit_ia32_PsiCondCMov(ia32_emit_env_t *env, const ir_node *node) {
+       CMov_emitter(env, node);
 }
 
-static void emit_ia32_xCmpCMov(ir_node *irn, ia32_emit_env_t *env) {
-       CMov_emitter(irn, env);
+static
+void emit_ia32_xCmpCMov(ia32_emit_env_t *env, const ir_node *node) {
+       CMov_emitter(env, node);
 }
 
-static void Set_emitter(ir_node *irn, ia32_emit_env_t *env) {
-       FILE               *F       = env->out;
-       const lc_arg_env_t *arg_env = ia32_get_arg_env();
-       const char *cmp_suffix = get_cmp_suffix(get_ia32_pncode(irn), ! mode_is_signed(get_irn_mode(get_irn_n(irn, 0))));
-       const char *instr      = "xor";
+static
+void Set_emitter(ia32_emit_env_t *env, const ir_node *node, ir_mode *mode) {
+       int pnc = get_ia32_pncode(node);
        const char *reg8bit;
-
-       char cmd_buf[SNPRINTF_BUF_LEN];
-       char cmnt_buf[SNPRINTF_BUF_LEN];
        const arch_register_t *out;
 
-       out     = arch_get_irn_register(env->arch_env, irn);
+       out     = arch_get_irn_register(env->arch_env, node);
        reg8bit = ia32_get_mapped_reg_name(env->isa->regs_8bit, out);
 
-       if (env->isa->opt_arch == arch_pentium_4) {
-               /* P4 prefers sub r, r, others xor r, r */
-               instr = "sub";
-       }
-
-       /* in case of a PsiCondSet use mov because it doesn't affect the eflags */
-       if (is_ia32_PsiCondSet(irn)) {
-               snprintf(cmd_buf, SNPRINTF_BUF_LEN, "mov %%%s, 0", arch_register_get_name(out));
-       }
-       else {
-               snprintf(cmd_buf, SNPRINTF_BUF_LEN, "%s %%%s, %%%s", instr, arch_register_get_name(out), arch_register_get_name(out));
-       }
-
-       snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* clear target as set modifies only lower 8 bit */");
-       IA32_DO_EMIT(irn);
-
-       if (is_ia32_CmpSet(irn)) {
-               lc_esnprintf(arg_env, cmd_buf, SNPRINTF_BUF_LEN, "cmp %s", ia32_emit_binop(irn, env));
-       }
-       else if (is_ia32_xCmpSet(irn)) {
-               lc_esnprintf(arg_env, cmd_buf, SNPRINTF_BUF_LEN, "ucomis%M %s", get_irn_n(irn, 0), ia32_emit_binop(irn, env));
-       }
-       else if (is_ia32_PsiCondSet(irn)) {
-               /* omit compare because flags are already set by And/Or */
-               snprintf(cmd_buf, SNPRINTF_BUF_LEN, " ");
-       }
-       else {
+       if (is_ia32_CmpSet(node)) {
+               be_emit_cstring(env, "\tcmp ");
+               ia32_emit_binop(env, node);
+       } else if (is_ia32_xCmpSet(node)) {
+               be_emit_cstring(env, "\tucomis");
+               ia32_emit_mode_suffix(env, get_irn_mode(get_irn_n(node, 2)));
+               be_emit_char(env, ' ');
+               ia32_emit_binop(env, node);
+       } else if (is_ia32_PsiCondSet(node)) {
+               be_emit_cstring(env, "\tcmp $0, ");
+               ia32_emit_source_register(env, node, 0);
+       } else {
                assert(0 && "unsupported Set");
        }
-       snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* calculate Psi condition */" );
-       IA32_DO_EMIT(irn);
+       be_emit_finish_line_gas(env, node);
+
+       /* use mov to clear target because it doesn't affect the eflags */
+       be_emit_cstring(env, "\tmovl $0, %");
+       be_emit_string(env, arch_register_get_name(out));
+       be_emit_finish_line_gas(env, node);
 
-       snprintf(cmd_buf, SNPRINTF_BUF_LEN, "set%s %%%s", cmp_suffix, reg8bit);
-       snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* set 1 iff true, 0 otherweise */" );
-       IA32_DO_EMIT(irn);
+       be_emit_cstring(env, "\tset");
+       ia32_emit_cmp_suffix(env, pnc);
+       be_emit_cstring(env, " %");
+       be_emit_string(env, reg8bit);
+       be_emit_finish_line_gas(env, node);
 }
 
-static void emit_ia32_CmpSet(ir_node *irn, ia32_emit_env_t *env) {
-       Set_emitter(irn, env);
+static
+void emit_ia32_CmpSet(ia32_emit_env_t *env, const ir_node *node) {
+       Set_emitter(env, node, get_irn_mode(get_irn_n(node, 2)));
 }
 
-static void emit_ia32_PsiCondSet(ir_node *irn, ia32_emit_env_t *env) {
-       Set_emitter(irn, env);
+static
+void emit_ia32_PsiCondSet(ia32_emit_env_t *env, const ir_node *node) {
+       Set_emitter(env, node, get_irn_mode(get_irn_n(node, 0)));
 }
 
-static void emit_ia32_xCmpSet(ir_node *irn, ia32_emit_env_t *env) {
-       Set_emitter(irn, env);
+static
+void emit_ia32_xCmpSet(ia32_emit_env_t *env, const ir_node *node) {
+       Set_emitter(env, node, get_irn_mode(get_irn_n(node, 2)));
 }
 
-static void emit_ia32_xCmp(ir_node *irn, ia32_emit_env_t *env) {
-       FILE               *F       = env->out;
-       const lc_arg_env_t *arg_env = ia32_get_arg_env();
-       int                sse_pnc  = -1;
-       char cmd_buf[SNPRINTF_BUF_LEN];
-       char cmnt_buf[SNPRINTF_BUF_LEN];
+static
+void emit_ia32_xCmp(ia32_emit_env_t *env, const ir_node *node) {
+       int  sse_pnc  = -1;
+       long pnc      = get_ia32_pncode(node);
+       long unord    = pnc & pn_Cmp_Uo;
 
-       switch (get_ia32_pncode(irn)) {
+       assert( (pnc & ia32_pn_Cmp_Unsigned) == 0);
+
+       switch (pnc) {
                case pn_Cmp_Leg: /* odered */
                        sse_pnc = 7;
                        break;
                case pn_Cmp_Uo:  /* unordered */
                        sse_pnc = 3;
                        break;
-               case pn_Cmp_Ue:  /* == */
+               case pn_Cmp_Ue:
+               case pn_Cmp_Eq:  /* == */
                        sse_pnc = 0;
                        break;
-               case pn_Cmp_Ul:  /* < */
+               case pn_Cmp_Ul:
+               case pn_Cmp_Lt:  /* < */
                        sse_pnc = 1;
                        break;
-               case pn_Cmp_Ule: /* <= */
+               case pn_Cmp_Ule:
+               case pn_Cmp_Le: /* <= */
                        sse_pnc = 2;
                        break;
-               case pn_Cmp_Ug:  /* > */
+               case pn_Cmp_Ug:
+               case pn_Cmp_Gt:  /* > */
                        sse_pnc = 6;
                        break;
-               case pn_Cmp_Uge: /* >= */
+               case pn_Cmp_Uge:
+               case pn_Cmp_Ge: /* >= */
                        sse_pnc = 5;
                        break;
-               case pn_Cmp_Ne:  /* != */
+               case pn_Cmp_Ne:
+               case pn_Cmp_Lg:  /* != */
                        sse_pnc = 4;
                        break;
        }
 
-       assert(sse_pnc >= 0 && "unsupported floating point compare");
+       assert(sse_pnc >= 0 && "unsupported compare");
+
+       if (unord && sse_pnc != 3) {
+               /*
+                       We need a separate compare against unordered.
+                       Quick and Dirty solution:
+                       - get some memory on stack
+                       - compare
+                       - store result
+                       - compare
+                       - and result and stored result
+                   - cleanup stack
+               */
+               be_emit_cstring(env, "\tsubl $8, %esp");
+               be_emit_finish_line_gas(env, node);
+
+               be_emit_cstring(env, "\tcmpsd $3, ");
+               ia32_emit_binop(env, node);
+               be_emit_finish_line_gas(env, node);
+
+               be_emit_cstring(env, "\tmovsd ");
+               ia32_emit_dest_register(env, node, 0);
+               be_emit_cstring(env, ", (%esp)");
+               be_emit_finish_line_gas(env, node);
+       }
+
+       be_emit_cstring(env, "\tcmpsd ");
+       be_emit_irprintf(env->emit, "%d, ", sse_pnc);
+       ia32_emit_binop(env, node);
+       be_emit_finish_line_gas(env, node);
+
+       if (unord && sse_pnc != 3) {
+               be_emit_cstring(env, "\tandpd (%esp), ");
+               ia32_emit_dest_register(env, node, 0);
+               be_emit_finish_line_gas(env, node);
 
-       lc_esnprintf(arg_env, cmd_buf, SNPRINTF_BUF_LEN, "cmps%M %s, %d", irn, ia32_emit_binop(irn, env), sse_pnc);
-       lc_esnprintf(arg_env, cmnt_buf, SNPRINTF_BUF_LEN, "/* SSE compare with result in %1D */", irn);
-       IA32_DO_EMIT(irn);
+               be_emit_cstring(env, "\taddl $8, %esp");
+               be_emit_finish_line_gas(env, node);
+       }
 }
 
 /*********************************************************
@@ -1195,7 +1137,8 @@ typedef struct _jmp_tbl_t {
 /**
  * Compare two variables of type branch_t. Used to sort all switch cases
  */
-static int ia32_cmp_branch_t(const void *a, const void *b) {
+static
+int ia32_cmp_branch_t(const void *a, const void *b) {
        branch_t *b1 = (branch_t *)a;
        branch_t *b2 = (branch_t *)b;
 
@@ -1210,43 +1153,41 @@ static int ia32_cmp_branch_t(const void *a, const void *b) {
  * possible otherwise a cmp-jmp cascade). Port from
  * cggg ia32 backend
  */
-static void emit_ia32_SwitchJmp(const ir_node *irn, ia32_emit_env_t *emit_env) {
+static
+void emit_ia32_SwitchJmp(ia32_emit_env_t *env, const ir_node *node) {
        unsigned long       interval;
-       char                buf[SNPRINTF_BUF_LEN];
-       int                 last_value, i, pn;
+       int                 last_value, i;
+       long                pnc;
        jmp_tbl_t           tbl;
        ir_node            *proj;
        const ir_edge_t    *edge;
-       const lc_arg_env_t *env = ia32_get_arg_env();
-       FILE               *F   = emit_env->out;
-       char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
 
        /* fill the table structure */
        tbl.label        = xmalloc(SNPRINTF_BUF_LEN);
-       tbl.label        = get_unique_label(tbl.label, SNPRINTF_BUF_LEN, "JMPTBL_");
+       tbl.label        = get_unique_label(tbl.label, SNPRINTF_BUF_LEN, ".TBL_");
        tbl.defProj      = NULL;
-       tbl.num_branches = get_irn_n_edges(irn);
+       tbl.num_branches = get_irn_n_edges(node);
        tbl.branches     = xcalloc(tbl.num_branches, sizeof(tbl.branches[0]));
        tbl.min_value    = INT_MAX;
        tbl.max_value    = INT_MIN;
 
        i = 0;
        /* go over all proj's and collect them */
-       foreach_out_edge(irn, edge) {
+       foreach_out_edge(node, edge) {
                proj = get_edge_src_irn(edge);
                assert(is_Proj(proj) && "Only proj allowed at SwitchJmp");
 
-               pn = get_Proj_proj(proj);
+               pnc = get_Proj_proj(proj);
 
                /* create branch entry */
                tbl.branches[i].target = proj;
-               tbl.branches[i].value  = pn;
+               tbl.branches[i].value  = pnc;
 
-               tbl.min_value = pn < tbl.min_value ? pn : tbl.min_value;
-               tbl.max_value = pn > tbl.max_value ? pn : tbl.max_value;
+               tbl.min_value = pnc < tbl.min_value ? pnc : tbl.min_value;
+               tbl.max_value = pnc > tbl.max_value ? pnc : tbl.max_value;
 
                /* check for default proj */
-               if (pn == get_ia32_pncode(irn)) {
+               if (pnc == get_ia32_pncode(node)) {
                        assert(tbl.defProj == NULL && "found two defProjs at SwitchJmp");
                        tbl.defProj = proj;
                }
@@ -1261,48 +1202,53 @@ static void emit_ia32_SwitchJmp(const ir_node *irn, ia32_emit_env_t *emit_env) {
        interval = tbl.max_value - tbl.min_value;
 
        /* emit the table */
-       lc_esnprintf(env, cmd_buf, SNPRINTF_BUF_LEN, "cmp %1S, %u", irn, interval);
-       snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* compare for switch */");
-       IA32_DO_EMIT(irn);
+       be_emit_cstring(env, "\tcmpl $");
+       be_emit_irprintf(env->emit, "%u, ", interval);
+       ia32_emit_source_register(env, node, 0);
+       be_emit_finish_line_gas(env, node);
 
-       snprintf(cmd_buf, SNPRINTF_BUF_LEN, "ja %s", get_cfop_target(tbl.defProj, buf));
-       snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* default jump if out of range  */");
-       IA32_DO_EMIT(irn);
+       be_emit_cstring(env, "\tja ");
+       ia32_emit_cfop_target(env, tbl.defProj);
+       be_emit_finish_line_gas(env, node);
 
        if (tbl.num_branches > 1) {
                /* create table */
+               be_emit_cstring(env, "\tjmp *");
+               be_emit_string(env, tbl.label);
+               be_emit_cstring(env, "(,");
+               ia32_emit_source_register(env, node, 0);
+               be_emit_cstring(env, ",4)");
+               be_emit_finish_line_gas(env, node);
 
-               lc_esnprintf(env, cmd_buf, SNPRINTF_BUF_LEN, "jmp %s[%1S*4]", tbl.label, irn);
-               snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* get jump table entry as target */");
-               IA32_DO_EMIT(irn);
-
-               ia32_switch_section(F, SECTION_RODATA);
-               fprintf(F, "\t.align 4\n");
+               be_gas_emit_switch_section(env->emit, GAS_SECTION_RODATA);
+               be_emit_cstring(env, "\t.align 4\n");
+               be_emit_write_line(env);
 
-               fprintf(F, "%s:\n", tbl.label);
+               be_emit_string(env, tbl.label);
+               be_emit_cstring(env, ":\n");
+               be_emit_write_line(env);
 
-               snprintf(cmd_buf, SNPRINTF_BUF_LEN, ".long %s", get_cfop_target(tbl.branches[0].target, buf));
-               snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* case %d */",  tbl.branches[0].value);
-               IA32_DO_EMIT(irn);
+               be_emit_cstring(env, ".long ");
+               ia32_emit_cfop_target(env, tbl.branches[0].target);
+               be_emit_finish_line_gas(env, NULL);
 
                last_value = tbl.branches[0].value;
                for (i = 1; i < tbl.num_branches; ++i) {
                        while (++last_value < tbl.branches[i].value) {
-                               snprintf(cmd_buf, SNPRINTF_BUF_LEN, ".long %s", get_cfop_target(tbl.defProj, buf));
-                               snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* default case */");
-                               IA32_DO_EMIT(irn);
+                               be_emit_cstring(env, ".long ");
+                               ia32_emit_cfop_target(env, tbl.defProj);
+                               be_emit_finish_line_gas(env, NULL);
                        }
-                       snprintf(cmd_buf, SNPRINTF_BUF_LEN, ".long %s", get_cfop_target(tbl.branches[i].target, buf));
-                       snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* case %d */", last_value);
-                       IA32_DO_EMIT(irn);
+                       be_emit_cstring(env, ".long ");
+                       ia32_emit_cfop_target(env, tbl.branches[i].target);
+                       be_emit_finish_line_gas(env, NULL);
                }
-               ia32_switch_section(F, SECTION_TEXT);
-       }
-       else {
+               be_gas_emit_switch_section(env->emit, GAS_SECTION_TEXT);
+       } else {
                /* one jump is enough */
-               snprintf(cmd_buf, SNPRINTF_BUF_LEN, "jmp %s", get_cfop_target(tbl.branches[0].target, buf));
-               snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* only one case given */");
-               IA32_DO_EMIT(irn);
+               be_emit_cstring(env, "\tjmp ");
+               ia32_emit_cfop_target(env, tbl.branches[0].target);
+               be_emit_finish_line_gas(env, node);
        }
 
        if (tbl.label)
@@ -1314,53 +1260,24 @@ static void emit_ia32_SwitchJmp(const ir_node *irn, ia32_emit_env_t *emit_env) {
 /**
  * Emits code for a unconditional jump.
  */
-static void emit_Jmp(const ir_node *irn, ia32_emit_env_t *env) {
-       ir_node *block, *next_bl;
-       FILE *F = env->out;
-       char buf[SNPRINTF_BUF_LEN], cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
+static
+void emit_Jmp(ia32_emit_env_t *env, const ir_node *node) {
+       ir_node *block, *next_block;
 
        /* for now, the code works for scheduled and non-schedules blocks */
-       block = get_nodes_block(irn);
+       block = get_nodes_block(node);
 
        /* we have a block schedule */
-       next_bl = next_blk_sched(block);
-       if (get_cfop_target_block(irn) != next_bl) {
-               snprintf(cmd_buf, SNPRINTF_BUF_LEN, "jmp %s", get_cfop_target(irn, buf));
-               lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F(%+F) */", irn, get_cfop_target_block(irn));
-       }
-       else {
-               cmd_buf[0] = '\0';
-               lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* fallthrough %s */", get_cfop_target(irn, buf));
-       }
-       IA32_DO_EMIT(irn);
-}
-
-/****************************
- *                  _
- *                 (_)
- *  _ __  _ __ ___  _  ___
- * | '_ \| '__/ _ \| |/ __|
- * | |_) | | | (_) | |\__ \
- * | .__/|_|  \___/| ||___/
- * | |            _/ |
- * |_|           |__/
- ****************************/
-
-/**
- * Emits code for a proj -> node
- */
-static void emit_Proj(const ir_node *irn, ia32_emit_env_t *env) {
-       ir_node *pred = get_Proj_pred(irn);
-
-       if (get_irn_op(pred) == op_Start) {
-               switch(get_Proj_proj(irn)) {
-                       case pn_Start_X_initial_exec:
-                               emit_Jmp(irn, env);
-                               break;
-                       default:
-                               break;
-               }
+       next_block = next_blk_sched(block);
+       if (get_cfop_target_block(node) != next_block) {
+               be_emit_cstring(env, "\tjmp ");
+               ia32_emit_cfop_target(env, node);
+       } else {
+               be_emit_cstring(env, "\t/* fallthrough to ");
+               ia32_emit_cfop_target(env, node);
+               be_emit_cstring(env, " */");
        }
+       be_emit_finish_line_gas(env, node);
 }
 
 /**********************************
@@ -1377,76 +1294,57 @@ static void emit_Proj(const ir_node *irn, ia32_emit_env_t *env) {
 /**
  * Emit movsb/w instructions to make mov count divideable by 4
  */
-static void emit_CopyB_prolog(FILE *F, int rem, int size) {
-       char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
-
-       fprintf(F, "\t/* memcopy %d bytes */\n", size);
-
-       snprintf(cmd_buf, SNPRINTF_BUF_LEN, "cld");
-       snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* copy direction forward */");
+static
+void emit_CopyB_prolog(ia32_emit_env_t *env, int rem) {
+       be_emit_cstring(env, "\tcld");
+       be_emit_finish_line_gas(env, NULL);
 
        switch(rem) {
-               case 1:
-                       IA32_DO_EMIT(NULL);
-                       snprintf(cmd_buf, SNPRINTF_BUF_LEN, "movsb");
-                       snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* memcopy remainder 1 */");
-                       break;
-               case 2:
-                       IA32_DO_EMIT(NULL);
-                       snprintf(cmd_buf, SNPRINTF_BUF_LEN, "movsw");
-                       snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* memcopy remainder 2 */");
-                       break;
-               case 3:
-                       IA32_DO_EMIT(NULL);
-                       snprintf(cmd_buf, SNPRINTF_BUF_LEN, "movsb");
-                       snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* memcopy remainder 3 */");
-                       IA32_DO_EMIT(NULL);
-                       snprintf(cmd_buf, SNPRINTF_BUF_LEN, "movsw");
-                       snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* memcopy remainder 3 */");
-                       break;
+       case 1:
+               be_emit_cstring(env, "\tmovsb");
+               be_emit_finish_line_gas(env, NULL);
+               break;
+       case 2:
+               be_emit_cstring(env, "\tmovsw");
+               be_emit_finish_line_gas(env, NULL);
+               break;
+       case 3:
+               be_emit_cstring(env, "\tmovsb");
+               be_emit_finish_line_gas(env, NULL);
+               be_emit_cstring(env, "\tmovsw");
+               be_emit_finish_line_gas(env, NULL);
+               break;
        }
-
-       IA32_DO_EMIT(NULL);
 }
 
 /**
  * Emit rep movsd instruction for memcopy.
  */
-static void emit_ia32_CopyB(const ir_node *irn, ia32_emit_env_t *emit_env) {
-       FILE    *F         = emit_env->out;
-       tarval  *tv        = get_ia32_Immop_tarval(irn);
-       int      rem       = get_tarval_long(tv);
-       ir_node *size_node = get_irn_n(irn, 2);
-       int      size;
-       char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
+static
+void emit_ia32_CopyB(ia32_emit_env_t *env, const ir_node *node) {
+       tarval *tv = get_ia32_Immop_tarval(node);
+       int    rem = get_tarval_long(tv);
 
-       /* beware: size_node could be a be_Copy to fulfill constraints for ecx */
-       size_node = be_is_Copy(size_node) ? be_get_Copy_op(size_node) : size_node;
-       size      = get_tarval_long(get_ia32_Immop_tarval(size_node));
+       emit_CopyB_prolog(env, rem);
 
-       emit_CopyB_prolog(F, rem, size);
-
-       snprintf(cmd_buf, SNPRINTF_BUF_LEN, "rep movsd");
-       snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* memcopy */");
-       IA32_DO_EMIT(irn);
+       be_emit_cstring(env, "\trep movsd");
+       be_emit_finish_line_gas(env, node);
 }
 
 /**
  * Emits unrolled memcopy.
  */
-static void emit_ia32_CopyB_i(const ir_node *irn, ia32_emit_env_t *emit_env) {
-       tarval *tv   = get_ia32_Immop_tarval(irn);
+static
+void emit_ia32_CopyB_i(ia32_emit_env_t *env, const ir_node *node) {
+       tarval *tv   = get_ia32_Immop_tarval(node);
        int     size = get_tarval_long(tv);
-       FILE   *F    = emit_env->out;
-       char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
 
-       emit_CopyB_prolog(F, size & 0x3, size);
+       emit_CopyB_prolog(env, size & 0x3);
 
        size >>= 2;
        while (size--) {
-               snprintf(cmd_buf, SNPRINTF_BUF_LEN, "movsd");
-               snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* memcopy unrolled */");
-               IA32_DO_EMIT(irn);
+               be_emit_cstring(env, "\tmovsd");
+               be_emit_finish_line_gas(env, NULL);
        }
 }
 
@@ -1465,119 +1363,148 @@ static void emit_ia32_CopyB_i(const ir_node *irn, ia32_emit_env_t *emit_env) {
 /**
  * Emit code for conversions (I, FP), (FP, I) and (FP, FP).
  */
-static void emit_ia32_Conv_with_FP(const ir_node *irn, ia32_emit_env_t *emit_env) {
-       FILE               *F        = emit_env->out;
-       const lc_arg_env_t *env      = ia32_get_arg_env();
-       ir_mode            *src_mode = get_ia32_src_mode(irn);
-       ir_mode            *tgt_mode = get_ia32_tgt_mode(irn);
-       char               *from, *to, buf[64];
-       char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
-
-       from = mode_is_float(src_mode) ? (get_mode_size_bits(src_mode) == 32 ? "ss" : "sd") : "si";
-       to   = mode_is_float(tgt_mode) ? (get_mode_size_bits(tgt_mode) == 32 ? "ss" : "sd") : "si";
-
-       switch(get_ia32_op_type(irn)) {
+static
+void emit_ia32_Conv_with_FP(ia32_emit_env_t *env, const ir_node *node) {
+       ir_mode            *ls_mode = get_ia32_ls_mode(node);
+       int                 ls_bits = get_mode_size_bits(ls_mode);
+
+       be_emit_cstring(env, "\tcvt");
+
+       if(is_ia32_Conv_I2FP(node)) {
+               if(ls_bits == 32) {
+                       be_emit_cstring(env, "si2ss");
+               } else {
+                       be_emit_cstring(env, "si2sd");
+               }
+       } else if(is_ia32_Conv_FP2I(node)) {
+               if(ls_bits == 32) {
+                       be_emit_cstring(env, "ss2si");
+               } else {
+                       be_emit_cstring(env, "sd2si");
+               }
+       } else {
+               assert(is_ia32_Conv_FP2FP(node));
+               if(ls_bits == 32) {
+                       be_emit_cstring(env, "sd2ss");
+               } else {
+                       be_emit_cstring(env, "ss2sd");
+               }
+       }
+       be_emit_char(env, ' ');
+
+       switch(get_ia32_op_type(node)) {
                case ia32_Normal:
-                       lc_esnprintf(env, buf, sizeof(buf), "%1D, %3S", irn, irn);
+                       ia32_emit_source_register(env, node, 2);
+                       be_emit_cstring(env, ", ");
+                       ia32_emit_dest_register(env, node, 0);
                        break;
                case ia32_AddrModeS:
-                       lc_esnprintf(env, buf, sizeof(buf), "%1D, %s", irn, ia32_emit_am(irn, emit_env));
+                       ia32_emit_dest_register(env, node, 0);
+                       be_emit_cstring(env, ", ");
+                       ia32_emit_am(env, node);
                        break;
                default:
                        assert(0 && "unsupported op type for Conv");
        }
-
-       snprintf(cmd_buf, SNPRINTF_BUF_LEN, "cvt%s2%s %s", from, to, buf);
-       lc_esnprintf(env, cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F(%+F, %+F) */", irn, src_mode, tgt_mode);
-       IA32_DO_EMIT(irn);
+       be_emit_finish_line_gas(env, node);
 }
 
-static void emit_ia32_Conv_I2FP(const ir_node *irn, ia32_emit_env_t *emit_env) {
-       emit_ia32_Conv_with_FP(irn, emit_env);
+static
+void emit_ia32_Conv_I2FP(ia32_emit_env_t *env, const ir_node *node) {
+       emit_ia32_Conv_with_FP(env, node);
 }
 
-static void emit_ia32_Conv_FP2I(const ir_node *irn, ia32_emit_env_t *emit_env) {
-       emit_ia32_Conv_with_FP(irn, emit_env);
+static
+void emit_ia32_Conv_FP2I(ia32_emit_env_t *env, const ir_node *node) {
+       emit_ia32_Conv_with_FP(env, node);
 }
 
-static void emit_ia32_Conv_FP2FP(const ir_node *irn, ia32_emit_env_t *emit_env) {
-       emit_ia32_Conv_with_FP(irn, emit_env);
+static
+void emit_ia32_Conv_FP2FP(ia32_emit_env_t *env, const ir_node *node) {
+       emit_ia32_Conv_with_FP(env, node);
 }
 
 /**
  * Emits code for an Int conversion.
  */
-static void emit_ia32_Conv_I2I(const ir_node *irn, ia32_emit_env_t *emit_env) {
-       FILE               *F        = emit_env->out;
-       const lc_arg_env_t *env      = ia32_get_arg_env();
-       char               *move_cmd = "movzx";
-       char               *conv_cmd = NULL;
-       ir_mode            *src_mode = get_ia32_src_mode(irn);
-       ir_mode            *tgt_mode = get_ia32_tgt_mode(irn);
-       int n, m;
-       char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
+static
+void emit_ia32_Conv_I2I(ia32_emit_env_t *env, const ir_node *node) {
+       const char *sign_suffix;
+       ir_mode *smaller_mode = get_ia32_ls_mode(node);
+       int smaller_bits = get_mode_size_bits(smaller_mode);
+       int signed_mode;
        const arch_register_t *in_reg, *out_reg;
 
-       n = get_mode_size_bits(src_mode);
-       m = get_mode_size_bits(tgt_mode);
-
-       if (mode_is_signed(n < m ? src_mode : tgt_mode)) {
-               move_cmd = "movsx";
-               if (n == 8 || m == 8)
-                       conv_cmd = "cbw";
-               else if (n == 16 || m == 16)
-                       conv_cmd = "cwde";
-               else
-                       assert(0 && "unsupported Conv_I2I");
+       assert(!mode_is_float(smaller_mode));
+       assert(smaller_bits == 8 || smaller_bits == 16 || smaller_bits == 32);
+
+       signed_mode = mode_is_signed(smaller_mode);
+       if(smaller_bits == 32) {
+               // this should not happen as it's no convert
+               assert(0);
+               sign_suffix = "";
+       } else {
+               sign_suffix = signed_mode ? "s" : "z";
        }
 
-       switch(get_ia32_op_type(irn)) {
+       switch(get_ia32_op_type(node)) {
                case ia32_Normal:
-                       in_reg  = get_in_reg(irn, 2);
-                       out_reg = get_out_reg(irn, 0);
+                       in_reg  = get_in_reg(env, node, 2);
+                       out_reg = get_out_reg(env, node, 0);
 
                        if (REGS_ARE_EQUAL(in_reg, &ia32_gp_regs[REG_EAX]) &&
                                REGS_ARE_EQUAL(out_reg, in_reg)                &&
-                               mode_is_signed(n < m ? src_mode : tgt_mode))
+                               signed_mode)
                        {
                                /* argument and result are both in EAX and */
                                /* signedness is ok: -> use converts       */
-                               lc_esnprintf(env, cmd_buf, SNPRINTF_BUF_LEN, "%s", conv_cmd);
-                       }
-                       else if (REGS_ARE_EQUAL(out_reg, in_reg) &&
-                               ! mode_is_signed(n < m ? src_mode : tgt_mode))
-                       {
+                               if (smaller_bits == 8) {
+                                       be_emit_cstring(env, "\tcbtw");
+                               } else if (smaller_bits == 16) {
+                                       be_emit_cstring(env, "\tcwtl");
+                               } else {
+                                       assert(0);
+                               }
+                       } else if (REGS_ARE_EQUAL(out_reg, in_reg) && !signed_mode) {
                                /* argument and result are in the same register */
                                /* and signedness is ok: -> use and with mask   */
-                               int mask = (1 << (n < m ? n : m)) - 1;
-                               lc_esnprintf(env, cmd_buf, SNPRINTF_BUF_LEN, "and %1D, 0x%x", irn, mask);
+                               int mask = (1 << smaller_bits) - 1;
+                               be_emit_cstring(env, "\tandl $0x");
+                               be_emit_irprintf(env->emit, "%x, ", mask);
+                               ia32_emit_dest_register(env, node, 0);
+                       } else {
+                               const char *sreg = ia32_get_reg_name_for_mode(env, smaller_mode, in_reg);
+
+                               be_emit_cstring(env, "\tmov");
+                               be_emit_string(env, sign_suffix);
+                               ia32_emit_mode_suffix(env, smaller_mode);
+                               be_emit_cstring(env, "l %");
+                               be_emit_string(env, sreg);
+                               be_emit_cstring(env, ", ");
+                               ia32_emit_dest_register(env, node, 0);
                        }
-                       else {
-                               /* use move w/o sign extension */
-                               lc_esnprintf(env, cmd_buf, SNPRINTF_BUF_LEN, "%s %1D, %%%s",
-                                       move_cmd, irn, ia32_get_reg_name_for_mode(emit_env, n < m ? src_mode : tgt_mode, in_reg));
-                       }
-
                        break;
-               case ia32_AddrModeS:
-                       lc_esnprintf(env, cmd_buf, SNPRINTF_BUF_LEN, "%s %1D, %s",
-                               move_cmd, irn, ia32_emit_am(irn, emit_env));
+               case ia32_AddrModeS: {
+                       be_emit_cstring(env, "\tmov");
+                       be_emit_string(env, sign_suffix);
+                       ia32_emit_mode_suffix(env, smaller_mode);
+                       be_emit_cstring(env, "l %");
+                       ia32_emit_am(env, node);
+                       be_emit_cstring(env, ", ");
+                       ia32_emit_dest_register(env, node, 0);
                        break;
+               }
                default:
                        assert(0 && "unsupported op type for Conv");
        }
-
-       lc_esnprintf(env, cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F(%d Bit mode_%F -> %d Bit mode_%F) */",
-               irn, n, src_mode, m, tgt_mode);
-
-       IA32_DO_EMIT(irn);
+       be_emit_finish_line_gas(env, node);
 }
 
 /**
  * Emits code for an 8Bit Int conversion.
  */
-void emit_ia32_Conv_I2I8Bit(const ir_node *irn, ia32_emit_env_t *emit_env) {
-       emit_ia32_Conv_I2I(irn, emit_env);
+void emit_ia32_Conv_I2I8Bit(ia32_emit_env_t *env, const ir_node *node) {
+       emit_ia32_Conv_I2I(env, node);
 }
 
 
@@ -1594,136 +1521,195 @@ void emit_ia32_Conv_I2I8Bit(const ir_node *irn, ia32_emit_env_t *emit_env) {
 /**
  * Emits a backend call
  */
-static void emit_be_Call(const ir_node *irn, ia32_emit_env_t *emit_env) {
-       FILE *F = emit_env->out;
-       entity *ent = be_Call_get_entity(irn);
-       char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
+static
+void emit_be_Call(ia32_emit_env_t *env, const ir_node *node) {
+       ir_entity *ent = be_Call_get_entity(node);
 
+       be_emit_cstring(env, "\tcall ");
        if (ent) {
-               snprintf(cmd_buf, SNPRINTF_BUF_LEN, "call %s", get_entity_ld_name(ent));
-       }
-       else {
-               lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "%1D", get_irn_n(irn, be_pos_Call_ptr));
+               mark_entity_visited(ent);
+               be_emit_string(env, get_entity_ld_name(ent));
+       } else {
+               be_emit_char(env, '*');
+               ia32_emit_dest_register(env, get_irn_n(node, be_pos_Call_ptr), 0);
        }
-
-       lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F (be_Call) */", irn);
-
-       IA32_DO_EMIT(irn);
+       be_emit_finish_line_gas(env, node);
 }
 
 /**
  * Emits code to increase stack pointer.
  */
-static void emit_be_IncSP(const ir_node *irn, ia32_emit_env_t *emit_env) {
-       FILE          *F    = emit_env->out;
-       unsigned       offs = be_get_IncSP_offset(irn);
-       be_stack_dir_t dir  = be_get_IncSP_direction(irn);
-       char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
-
-       if (offs) {
-               if (dir == be_stack_dir_expand)
-                       lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "sub %1S, %u", irn, offs);
-               else
-                       lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "add %1S, %u", irn, offs);
-               lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F (IncSP) */", irn);
-       }
-       else {
-               snprintf(cmd_buf, SNPRINTF_BUF_LEN, " ");
-               lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* omitted %+F (IncSP) with 0 */", irn);
-       }
+static
+void emit_be_IncSP(ia32_emit_env_t *env, const ir_node *node) {
+       int offs = be_get_IncSP_offset(node);
+
+       if (offs == 0)
+               return;
 
-       IA32_DO_EMIT(irn);
+       if (offs > 0) {
+               be_emit_cstring(env, "\tsubl $");
+               be_emit_irprintf(env->emit, "%u, ", offs);
+               ia32_emit_source_register(env, node, 0);
+       } else {
+               be_emit_cstring(env, "\taddl $");
+               be_emit_irprintf(env->emit, "%u, ", -offs);
+               ia32_emit_source_register(env, node, 0);
+       }
+       be_emit_finish_line_gas(env, node);
 }
 
 /**
  * Emits code to set stack pointer.
  */
-static void emit_be_SetSP(const ir_node *irn, ia32_emit_env_t *emit_env) {
-       FILE *F = emit_env->out;
-       char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
-
-       lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "mov %1D, %3S", irn, irn);
-       lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F (restore SP) */", irn);
-       IA32_DO_EMIT(irn);
+static
+void emit_be_SetSP(ia32_emit_env_t *env, const ir_node *node) {
+       be_emit_cstring(env, "\tmovl ");
+       ia32_emit_source_register(env, node, 2);
+       be_emit_cstring(env, ", ");
+       ia32_emit_dest_register(env, node, 0);
+       be_emit_finish_line_gas(env, node);
 }
 
 /**
  * Emits code for Copy/CopyKeep.
  */
-static void Copy_emitter(const ir_node *irn, ir_node *op, ia32_emit_env_t *emit_env) {
-       FILE             *F    = emit_env->out;
-       const arch_env_t *aenv = emit_env->arch_env;
-       char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
+static
+void Copy_emitter(ia32_emit_env_t *env, const ir_node *node, const ir_node *op)
+{
+       const arch_env_t *aenv = env->arch_env;
+       ir_mode *mode;
 
-       if (REGS_ARE_EQUAL(arch_get_irn_register(aenv, irn), arch_get_irn_register(aenv, op)) ||
-               be_is_unknown_reg(arch_get_irn_register(aenv, op)))
+       if (REGS_ARE_EQUAL(arch_get_irn_register(aenv, node), arch_get_irn_register(aenv, op)) ||
+               arch_register_type_is(arch_get_irn_register(aenv, op), virtual))
                return;
 
-       if (mode_is_float(get_irn_mode(irn)))
-               lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "movs%M %1D, %1S", irn, irn, irn);
-       else
-               lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "mov %1D, %1S", irn, irn);
-       lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F */", irn);
-       IA32_DO_EMIT(irn);
+       mode = get_irn_mode(node);
+       if (mode == mode_LLu) {
+               be_emit_cstring(env, "\tmovsd ");
+               ia32_emit_source_register(env, node, 0);
+               be_emit_cstring(env, ", ");
+               ia32_emit_dest_register(env, node, 0);
+       } else {
+               be_emit_cstring(env, "\tmovl ");
+               ia32_emit_source_register(env, node, 0);
+               be_emit_cstring(env, ", ");
+               ia32_emit_dest_register(env, node, 0);
+       }
+       be_emit_finish_line_gas(env, node);
 }
 
-static void emit_be_Copy(const ir_node *irn, ia32_emit_env_t *emit_env) {
-       Copy_emitter(irn, be_get_Copy_op(irn), emit_env);
+static
+void emit_be_Copy(ia32_emit_env_t *env, const ir_node *node) {
+       Copy_emitter(env, node, be_get_Copy_op(node));
 }
 
-static void emit_be_CopyKeep(const ir_node *irn, ia32_emit_env_t *emit_env) {
-       Copy_emitter(irn, be_get_CopyKeep_op(irn), emit_env);
+static
+void emit_be_CopyKeep(ia32_emit_env_t *env, const ir_node *node) {
+       Copy_emitter(env, node, be_get_CopyKeep_op(node));
 }
 
 /**
  * Emits code for exchange.
  */
-static void emit_be_Perm(const ir_node *irn, ia32_emit_env_t *emit_env) {
-       FILE *F = emit_env->out;
-       char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
-
-       lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "xchg %1S, %2S", irn, irn);
-       lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F(%1A, %2A) */", irn, irn, irn);
-       IA32_DO_EMIT(irn);
+static
+void emit_be_Perm(ia32_emit_env_t *env, const ir_node *node) {
+       const arch_register_t *in1, *in2;
+       const arch_register_class_t *cls1, *cls2;
+
+       in1 = arch_get_irn_register(env->arch_env, get_irn_n(node, 0));
+       in2 = arch_get_irn_register(env->arch_env, get_irn_n(node, 1));
+
+       cls1 = arch_register_get_class(in1);
+       cls2 = arch_register_get_class(in2);
+
+       assert(cls1 == cls2 && "Register class mismatch at Perm");
+
+       if (cls1 == &ia32_reg_classes[CLASS_ia32_gp]) {
+               be_emit_cstring(env, "\txchg ");
+               ia32_emit_source_register(env, node, 1);
+               be_emit_cstring(env, ", ");
+               ia32_emit_source_register(env, node, 0);
+               be_emit_finish_line_gas(env, node);
+       } else if (cls1 == &ia32_reg_classes[CLASS_ia32_xmm]) {
+               be_emit_cstring(env, "\txorpd ");
+               ia32_emit_source_register(env, node, 1);
+               be_emit_cstring(env, ", ");
+               ia32_emit_source_register(env, node, 0);
+               be_emit_finish_line_gas(env, NULL);
+
+               be_emit_cstring(env, "\txorpd ");
+               ia32_emit_source_register(env, node, 0);
+               be_emit_cstring(env, ", ");
+               ia32_emit_source_register(env, node, 1);
+               be_emit_finish_line_gas(env, NULL);
+
+               be_emit_cstring(env, "\txorpd ");
+               ia32_emit_source_register(env, node, 1);
+               be_emit_cstring(env, ", ");
+               ia32_emit_source_register(env, node, 0);
+               be_emit_finish_line_gas(env, node);
+       } else if (cls1 == &ia32_reg_classes[CLASS_ia32_vfp]) {
+               /* is a NOP */
+       } else if (cls1 == &ia32_reg_classes[CLASS_ia32_st]) {
+               /* is a NOP */
+       }
 }
 
 /**
  * Emits code for Constant loading.
  */
-static void emit_ia32_Const(const ir_node *n, ia32_emit_env_t *env) {
-  FILE *F = env->out;
-  char cmd_buf[256], cmnt_buf[256];
-  const lc_arg_env_t *arg_env = ia32_get_arg_env();
-
-  if (get_ia32_Immop_tarval(n) == get_tarval_null(get_irn_mode(n))) {
-               const char *instr = "xor";
-               if (env->isa->opt_arch == arch_pentium_4) {
-                       /* P4 prefers sub r, r, others xor r, r */
-                       instr = "sub";
-               }
-    lc_esnprintf(arg_env, cmd_buf, 256, "%s %1D, %1D ", instr, n, n);
-    lc_esnprintf(arg_env, cmnt_buf, 256, "/* optimized mov 0 to register */");
-  }
-  else {
-    if (get_ia32_op_type(n) == ia32_SymConst) {
-      lc_esnprintf(arg_env, cmd_buf, 256, "mov %1D, OFFSET FLAT:%C ", n, n);
-      lc_esnprintf(arg_env, cmnt_buf, 256, "/* Move address of SymConst into register */");
-    }
-               else {
-                               lc_esnprintf(arg_env, cmd_buf, 256, "mov %1D, %C ", n, n);
-                               lc_esnprintf(arg_env, cmnt_buf, 256, "/* Mov Const into register */");
+static
+void emit_ia32_Const(ia32_emit_env_t *env, const ir_node *node) {
+       ia32_immop_type_t imm_tp = get_ia32_immop_type(node);
+
+       if (imm_tp == ia32_ImmSymConst) {
+               be_emit_cstring(env, "\tmovl ");
+               ia32_emit_immediate(env, node);
+               be_emit_cstring(env, ", ");
+               ia32_emit_dest_register(env, node, 0);
+       } else {
+               tarval *tv = get_ia32_Immop_tarval(node);
+               assert(get_irn_mode(node) == mode_Iu);
+               /* beware: in some rare cases mode is mode_b which has no tarval_null() */
+               if (tarval_is_null(tv)) {
+                       if (env->isa->opt_arch == arch_pentium_4) {
+                               /* P4 prefers sub r, r, others xor r, r */
+                               be_emit_cstring(env, "\tsubl ");
+                       } else {
+                               be_emit_cstring(env, "\txorl ");
+                       }
+                       ia32_emit_dest_register(env, node, 0);
+                       be_emit_cstring(env, ", ");
+                       ia32_emit_dest_register(env, node, 0);
+               } else {
+                       be_emit_cstring(env, "\tmovl ");
+                       ia32_emit_immediate(env, node);
+                       be_emit_cstring(env, ", ");
+                       ia32_emit_dest_register(env, node, 0);
                }
-  }
-  lc_efprintf(arg_env, F, "\t%-35s %-60s /* %+F (%+G) */\n", cmd_buf, cmnt_buf, n, n);
+       }
+       be_emit_finish_line_gas(env, node);
 }
 
-static void emit_be_Return(const ir_node *n, ia32_emit_env_t *env) {
-  FILE *F = env->out;
-  const lc_arg_env_t *arg_env = ia32_get_arg_env();
+/**
+ * Emits code to load the TLS base
+ */
+static
+void emit_ia32_LdTls(ia32_emit_env_t *env, const ir_node *node) {
+       be_emit_cstring(env, "\tmovl %gs:0, ");
+       ia32_emit_dest_register(env, node, 0);
+       be_emit_finish_line_gas(env, node);
+}
 
-  lc_efprintf(arg_env, F, "\t%-35s %-60s /* %+F (%+G) */\n", "ret", "/* be_Return */", n, n);
+static
+void emit_be_Return(ia32_emit_env_t *env, const ir_node *node) {
+       be_emit_cstring(env, "\tret");
+       be_emit_finish_line_gas(env, node);
 }
 
+static
+void emit_Nothing(ia32_emit_env_t *env, const ir_node *node) {
+}
 
 
 /***********************************************************************************
@@ -1740,12 +1726,15 @@ static void emit_be_Return(const ir_node *n, ia32_emit_env_t *env) {
  * Enters the emitter functions for handled nodes into the generic
  * pointer of an opcode.
  */
-static void ia32_register_emitters(void) {
+static
+void ia32_register_emitters(void) {
 
 #define IA32_EMIT2(a,b) op_ia32_##a->ops.generic = (op_func)emit_ia32_##b
 #define IA32_EMIT(a)    IA32_EMIT2(a,a)
 #define EMIT(a)         op_##a->ops.generic = (op_func)emit_##a
+#define IGN(a)                 op_##a->ops.generic = (op_func)emit_Nothing
 #define BE_EMIT(a)      op_be_##a->ops.generic = (op_func)emit_be_##a
+#define BE_IGN(a)              op_be_##a->ops.generic = (op_func)emit_Nothing
 
        /* first clear the generic function pointer for all ops */
        clear_irp_opcodes_generic_func();
@@ -1771,6 +1760,7 @@ static void ia32_register_emitters(void) {
        IA32_EMIT(Conv_I2I);
        IA32_EMIT(Conv_I2I8Bit);
        IA32_EMIT(Const);
+       IA32_EMIT(LdTls);
        IA32_EMIT(xCmp);
        IA32_EMIT(xCmpSet);
        IA32_EMIT(xCmpCMov);
@@ -1791,156 +1781,304 @@ static void ia32_register_emitters(void) {
        BE_EMIT(Perm);
        BE_EMIT(Return);
 
+       BE_IGN(RegParams);
+       BE_IGN(Barrier);
+       BE_IGN(Keep);
+
        /* firm emitter */
        EMIT(Jmp);
-       EMIT(Proj);
+       IGN(Proj);
+       IGN(Phi);
+       IGN(Start);
 
 #undef BE_EMIT
 #undef EMIT
+#undef IGN
 #undef IA32_EMIT2
 #undef IA32_EMIT
 }
 
+static const char *last_name = NULL;
+static unsigned last_line = -1;
+static unsigned num = -1;
+
+/**
+ * Emit the debug support for node node.
+ */
+static
+void ia32_emit_dbg(ia32_emit_env_t *env, const ir_node *node) {
+       dbg_info *db = get_irn_dbg_info(node);
+       unsigned lineno;
+       const char *fname = be_retrieve_dbg_info(db, &lineno);
+
+       if (! env->cg->birg->main_env->options->stabs_debug_support)
+               return;
+
+       if (fname) {
+               if (last_name != fname) {
+                       last_line = -1;
+                       be_dbg_include_begin(env->cg->birg->main_env->db_handle, fname);
+                       last_name = fname;
+               }
+               if (last_line != lineno) {
+                       char name[64];
+
+                       snprintf(name, sizeof(name), ".LM%u", ++num);
+                       last_line = lineno;
+                       be_dbg_line(env->cg->birg->main_env->db_handle, lineno, name);
+                       be_emit_string(env, name);
+                       be_emit_cstring(env, ":\n");
+                       be_emit_write_line(env);
+               }
+       }
+}
+
+typedef void (*emit_func_ptr) (ia32_emit_env_t *, const ir_node *);
+
 /**
  * Emits code for a node.
  */
-static void ia32_emit_node(const ir_node *irn, void *env) {
-       ia32_emit_env_t   *emit_env = env;
-       FILE              *F        = emit_env->out;
-       ir_op             *op       = get_irn_op(irn);
-       DEBUG_ONLY(firm_dbg_module_t *mod = emit_env->mod;)
+static
+void ia32_emit_node(ia32_emit_env_t *env, const ir_node *node) {
+       ir_op *op = get_irn_op(node);
 
-       DBG((mod, LEVEL_1, "emitting code for %+F\n", irn));
+       DBG((dbg, LEVEL_1, "emitting code for %+F\n", node));
 
        if (op->ops.generic) {
-               void (*emit)(const ir_node *, void *) = (void (*)(const ir_node *, void *))op->ops.generic;
-               (*emit)(irn, env);
-       }
-       else {
-               ir_fprintf(F, "\t%35s /* %+F (%+G) */\n", " ", irn, irn);
+               emit_func_ptr func = (emit_func_ptr) op->ops.generic;
+               ia32_emit_dbg(env, node);
+               (*func) (env, node);
+       } else {
+               emit_Nothing(env, node);
+               ir_fprintf(stderr, "Warning: No emit handler for node %+F (%+G)\n", node, node);
        }
 }
 
 /**
  * Emits gas alignment directives
  */
-static void ia32_emit_alignment(FILE *F, unsigned align, unsigned skip) {
-       fprintf(F, "\t.p2align %u,,%u\n", align, skip);
+static
+void ia32_emit_alignment(ia32_emit_env_t *env, unsigned align, unsigned skip) {
+       be_emit_cstring(env, "\t.p2align ");
+       be_emit_irprintf(env->emit, "%u,,%u\n", align, skip);
+       be_emit_write_line(env);
 }
 
 /**
  * Emits gas alignment directives for Functions depended on cpu architecture.
  */
-static void ia32_emit_align_func(FILE *F, cpu_support cpu) {
-       unsigned align; unsigned maximum_skip;
-
-       /* gcc doesn't emit alignment for p4 ?*/
-    if (cpu == arch_pentium_4)
-               return;
+static
+void ia32_emit_align_func(ia32_emit_env_t *env, cpu_support cpu) {
+       unsigned align;
+       unsigned maximum_skip;
 
        switch (cpu) {
                case arch_i386:
-                       align = 2; maximum_skip = 3;
+                       align = 2;
                        break;
                case arch_i486:
-                       align = 4; maximum_skip = 15;
+                       align = 4;
                        break;
                case arch_k6:
-                       align = 5; maximum_skip = 31;
+                       align = 5;
                        break;
                default:
-                       align = 4; maximum_skip = 15;
+                       align = 4;
        }
-       ia32_emit_alignment(F, align, maximum_skip);
+       maximum_skip = (1 << align) - 1;
+       ia32_emit_alignment(env, align, maximum_skip);
 }
 
 /**
  * Emits gas alignment directives for Labels depended on cpu architecture.
  */
-static void ia32_emit_align_label(FILE *F, cpu_support cpu) {
+static
+void ia32_emit_align_label(ia32_emit_env_t *env, cpu_support cpu) {
        unsigned align; unsigned maximum_skip;
 
-       /* gcc doesn't emit alignment for p4 ?*/
-    if (cpu == arch_pentium_4)
-               return;
-
        switch (cpu) {
                case arch_i386:
-                       align = 2; maximum_skip = 3;
+                       align = 2;
                        break;
                case arch_i486:
-                       align = 4; maximum_skip = 15;
+                       align = 4;
                        break;
                case arch_k6:
-                       align = 5; maximum_skip = 7;
+                       align = 5;
                        break;
                default:
-                       align = 4; maximum_skip = 7;
+                       align = 4;
        }
-       ia32_emit_alignment(F, align, maximum_skip);
+       maximum_skip = (1 << align) - 1;
+       ia32_emit_alignment(env, align, maximum_skip);
+}
+
+static
+int is_first_loop_block(ia32_emit_env_t *env, ir_node *block, ir_node *prev_block) {
+       ir_exec_freq *exec_freq = env->cg->birg->exec_freq;
+       double block_freq, prev_freq;
+       static const double DELTA = .0001;
+       cpu_support cpu = env->isa->opt_arch;
+
+       if(exec_freq == NULL)
+               return 0;
+       if(cpu == arch_i386 || cpu == arch_i486)
+               return 0;
+
+       block_freq = get_block_execfreq(exec_freq, block);
+       prev_freq = get_block_execfreq(exec_freq, prev_block);
+
+       if(block_freq < DELTA || prev_freq < DELTA)
+               return 0;
+
+       block_freq /= prev_freq;
+
+       switch (cpu) {
+               case arch_athlon:
+               case arch_athlon_64:
+               case arch_k6:
+                       return block_freq > 3;
+               default:
+                       break;
+       }
+
+       return block_freq > 2;
 }
 
 /**
  * Walks over the nodes in a block connected by scheduling edges
  * and emits code for each node.
  */
-static void ia32_gen_block(ir_node *block, void *env) {
-       ia32_emit_env_t *emit_env = env;
-       const ir_node *irn;
-       int need_label = block != get_irg_start_block(get_irn_irg(block));
+static
+void ia32_gen_block(ia32_emit_env_t *env, ir_node *block, ir_node *last_block) {
+       ir_graph      *irg         = get_irn_irg(block);
+       ir_node       *start_block = get_irg_start_block(irg);
+       int           need_label   = 1;
+       const ir_node *node;
+       int           i;
 
-       if (! is_Block(block))
-               return;
+       assert(is_Block(block));
+
+       if (block == start_block)
+               need_label = 0;
+
+       if (need_label && get_irn_arity(block) == 1) {
+               ir_node *pred_block = get_Block_cfgpred_block(block, 0);
 
-       if (need_label && (emit_env->cg->opt & IA32_OPT_EXTBB)) {
-               /* if the extended block scheduler is used, only leader blocks need
-                  labels. */
-               need_label = (block == get_extbb_leader(get_nodes_extbb(block)));
+               if (pred_block == last_block && get_irn_n_edges_kind(pred_block, EDGE_KIND_BLOCK) <= 2)
+                       need_label = 0;
+       }
+
+       /* special case: if one of our cfg preds is a switch-jmp we need a label, */
+       /*               otherwise there might be jump table entries jumping to   */
+       /*               non-existent (omitted) labels                            */
+       for (i = get_Block_n_cfgpreds(block) - 1; i >= 0; --i) {
+               ir_node *pred = get_Block_cfgpred(block, i);
+
+               if (is_Proj(pred)) {
+                       assert(get_irn_mode(pred) == mode_X);
+                       if (is_ia32_SwitchJmp(get_Proj_pred(pred))) {
+                               need_label = 1;
+                               break;
+                       }
+               }
        }
 
        if (need_label) {
-               ia32_emit_align_label(emit_env->out, emit_env->isa->opt_arch);
-               fprintf(emit_env->out, BLOCK_PREFIX("%ld:\n"), get_irn_node_nr(block));
+               int i, arity;
+               int align = 1;
+               ir_exec_freq *exec_freq = env->cg->birg->exec_freq;
+
+               /* align the loop headers */
+               if (! is_first_loop_block(env, block, last_block)) {
+                       /* align blocks where the previous block has no fallthrough */
+                       arity = get_irn_arity(block);
+
+                       for (i = 0; i < arity; ++i) {
+                               ir_node *predblock = get_Block_cfgpred_block(block, i);
+
+                               if (predblock == last_block) {
+                                       align = 0;
+                                       break;
+                               }
+                       }
+               }
+
+               if (align)
+                       ia32_emit_align_label(env, env->isa->opt_arch);
+
+               be_emit_cstring(env, BLOCK_PREFIX);
+               be_emit_irprintf(env->emit, "%d:", get_irn_node_nr(block));
+               be_emit_pad_comment(env);
+               be_emit_cstring(env, "   /* preds:");
+
+               /* emit list of pred blocks in comment */
+               arity = get_irn_arity(block);
+               for (i = 0; i < arity; ++i) {
+                       ir_node *predblock = get_Block_cfgpred_block(block, i);
+                       be_emit_irprintf(env->emit, " %d", get_irn_node_nr(predblock));
+               }
+
+               if (exec_freq != NULL) {
+                       be_emit_irprintf(env->emit, " freq: %f", get_block_execfreq(exec_freq, block));
+               }
+               be_emit_cstring(env, " */\n");
+               be_emit_write_line(env);
        }
 
-       sched_foreach(block, irn) {
-               ia32_emit_node(irn, env);
+       /* emit the contents of the block */
+       ia32_emit_dbg(env, block);
+       sched_foreach(block, node) {
+               ia32_emit_node(env, node);
        }
 }
 
 /**
  * Emits code for function start.
  */
-static void ia32_emit_func_prolog(FILE *F, ir_graph *irg, cpu_support cpu) {
-       entity     *irg_ent  = get_irg_entity(irg);
+static
+void ia32_emit_func_prolog(ia32_emit_env_t *env, ir_graph *irg) {
+       ir_entity  *irg_ent  = get_irg_entity(irg);
        const char *irg_name = get_entity_ld_name(irg_ent);
+       cpu_support cpu      = env->isa->opt_arch;
+       const be_irg_t *birg = env->cg->birg;
 
-       fprintf(F, "\n");
-       ia32_switch_section(F, SECTION_TEXT);
-       ia32_emit_align_func(F, cpu);
+       be_emit_write_line(env);
+       be_gas_emit_switch_section(env->emit, GAS_SECTION_TEXT);
+       be_dbg_method_begin(birg->main_env->db_handle, irg_ent, be_abi_get_stack_layout(birg->abi));
+       ia32_emit_align_func(env, cpu);
        if (get_entity_visibility(irg_ent) == visibility_external_visible) {
-               fprintf(F, ".globl %s\n", irg_name);
+               be_emit_cstring(env, ".global ");
+               be_emit_string(env, irg_name);
+               be_emit_char(env, '\n');
+               be_emit_write_line(env);
        }
-       ia32_dump_function_object(F, irg_name);
-       fprintf(F, "%s:\n", irg_name);
+       ia32_emit_function_object(env, irg_name);
+       be_emit_string(env, irg_name);
+       be_emit_cstring(env, ":\n");
+       be_emit_write_line(env);
 }
 
 /**
  * Emits code for function end
  */
-static void ia32_emit_func_epilog(FILE *F, ir_graph *irg) {
+static
+void ia32_emit_func_epilog(ia32_emit_env_t *env, ir_graph *irg) {
        const char *irg_name = get_entity_ld_name(get_irg_entity(irg));
+       const be_irg_t *birg = env->cg->birg;
 
-       ia32_dump_function_size(F, irg_name);
-       fprintf(F, "\n");
+       ia32_emit_function_size(env, irg_name);
+       be_dbg_method_end(birg->main_env->db_handle);
+       be_emit_char(env, '\n');
+       be_emit_write_line(env);
 }
 
 /**
  * Block-walker:
  * Sets labels for control flow nodes (jump target)
- * TODO: Jump optimization
  */
-static void ia32_gen_labels(ir_node *block, void *env) {
+static
+void ia32_gen_labels(ir_node *block, void *data) {
        ir_node *pred;
        int n = get_Block_n_cfgpreds(block);
 
@@ -1953,45 +2091,40 @@ static void ia32_gen_labels(ir_node *block, void *env) {
 /**
  * Main driver. Emits the code for one routine.
  */
-void ia32_gen_routine(FILE *F, ir_graph *irg, const ia32_code_gen_t *cg) {
-       ia32_emit_env_t emit_env;
+void ia32_gen_routine(ia32_code_gen_t *cg, ir_graph *irg) {
+       ia32_emit_env_t env;
        ir_node *block;
+       ir_node *last_block = NULL;
+       int i, n;
 
-       emit_env.out      = F;
-       emit_env.arch_env = cg->arch_env;
-       emit_env.cg       = cg;
-       emit_env.isa      = (ia32_isa_t *)cg->arch_env->isa;
-       FIRM_DBG_REGISTER(emit_env.mod, "firm.be.ia32.emitter");
-
-       /* set the global arch_env (needed by print hooks) */
-       arch_env = cg->arch_env;
+       env.isa      = (ia32_isa_t *)cg->arch_env->isa;
+       env.emit     = &env.isa->emit;
+       env.arch_env = cg->arch_env;
+       env.cg       = cg;
 
        ia32_register_emitters();
 
-       ia32_emit_func_prolog(F, irg, emit_env.isa->opt_arch);
-       irg_block_walk_graph(irg, ia32_gen_labels, NULL, &emit_env);
+       ia32_emit_func_prolog(&env, irg);
+       irg_block_walk_graph(irg, ia32_gen_labels, NULL, &env);
 
-       if ((cg->opt & IA32_OPT_EXTBB) && cg->blk_sched) {
-               int i, n = ARR_LEN(cg->blk_sched);
+       n = ARR_LEN(cg->blk_sched);
+       for (i = 0; i < n;) {
+               ir_node *next_bl;
 
-               for (i = 0; i < n;) {
-                       ir_node *next_bl;
+               block   = cg->blk_sched[i];
+               ++i;
+               next_bl = i < n ? cg->blk_sched[i] : NULL;
 
-                       block   = cg->blk_sched[i];
-                       ++i;
-                       next_bl = i < n ? cg->blk_sched[i] : NULL;
-
-                       /* set here the link. the emitter expects to find the next block here */
-                       set_irn_link(block, next_bl);
-                       ia32_gen_block(block, &emit_env);
-               }
-       }
-       else {
-               /* "normal" block schedule: Note the get_next_block() returns the NUMBER of the block
-                  in the block schedule. As this number should NEVER be equal the next block,
-                  we does not need a clear block link here. */
-               irg_walk_blkwise_graph(irg, NULL, ia32_gen_block, &emit_env);
+               /* set here the link. the emitter expects to find the next block here */
+               set_irn_link(block, next_bl);
+               ia32_gen_block(&env, block, last_block);
+               last_block = block;
        }
 
-       ia32_emit_func_epilog(F, irg);
+       ia32_emit_func_epilog(&env, irg);
+}
+
+void ia32_init_emitter(void)
+{
+       FIRM_DBG_REGISTER(dbg, "firm.be.ia32.emitter");
 }